Status and Trends of Northern Spotted Owl Populations and Habitats

Home , Biscuit Fire, Fahrenheit (crater), Northern spotted owl

NORTHWEST FOREST PLAN The First 15 Years (1994–2008)

Raymond J. Davis, Katie M. Dugger, Shawne Mohoric, Louisa Evers, and William C. Aney

General Technical Report United States PNW-GTR-850 Forest Pacific Northwest Department of Service Research Station October 2011 Agriculture The Forest Service of the U.S. Department of Agriculture is dedicated to the principle of multiple use management of the Nation’s forest resources for sustained yields of wood, water, forage, wildlife, and recreation. Through forestry research, cooperation with the States and private forest owners, and management of the national forests and national grasslands, it strives—as directed by Congress—to provide increasingly greater service to a growing Nation. The U.S. Department of Agriculture (USDA) prohibits discrimination in all its programs and activities on the basis of race, color, national origin, age, disability, and where applicable, sex, marital status, familial status, parental status, religion, sexual orientation, genetic information, political beliefs, reprisal, or because all or part of an individual’s income is derived from any public assistance program. (Not all prohibited bases apply to all programs.) Persons with disabilities who require alternative means for communication of program information (Braille, large print, audiotape, etc.) should contact USDA’s TARGET Center at (202) 720-2600 (voice and TDD). To file a complaint of discrimination, write USDA, Director, Office of Civil Rights, Room 1400 Independence Avenue, SW, Washington, DC 20250-9410 or call (800) 795-3272 (voice) or (202) 720-6382 (TDD). USDA is an equal opportunity provider and employer.

Authors Raymond J. Davis, Northern Spotted Owl Monitoring Module Leader, Northwest Forest Plan Interagency Monitoring Program, U.S. Department of Agriculture, Forest Service, Pacific Northwest Region, 2900 NW Stewart Parkway, Roseburg, OR 97471. Katie M. Dugger, associate professor, Senior Research, Department of Fisheries and Wildlife, Oregon State University, 104 Nash Hall, Corvallis, OR 97331-3803. Shawne Mohoric, program manager, NWFP Interagency Monitoring Program, U.S. Department of Agricul- ture, Forest Service, Pacific Northwest Region, 333 SW First Avenue, Portland, OR 97204. Louisa Evers, fire ecologist, U.S. Department of the Interior, Bureau of Land Management, Oregon State Office, 333 SW First Avenue, Portland, OR 97204. William C. Aney, regional fuels specialist, U.S. Department of Agriculture, Forest Service, Pacific Northwest Region, 2517 SW Hailey Avenue, Pendleton, OR 97801.

Cover: Photo by Peter Carlson. Preface This report is one of a set of periodic reports produced by the Northwest Forest Plan (the Plan) interagency monitoring program. These reports attempt to answer questions about the effectiveness of the Plan using the latest monitoring methods and research results. The reports focus on establishing baseline information from 1994, when the Plan was approved, and reporting changes that have occurred since then. The series includes late-successional and old-growth forests, northern spotted owl (Strix occidentalis caurina) population and habitat, marbled murrelet (Brachyramphus marmoratus) population and habitat, watershed condition, government-to-government tribal relationships, socioeconomic conditions, and project implementation. These monitoring reports are also intended to identify potential issues and to recommend solutions for future adaptive management changes and, as noted in the first reporting cycle, to resolve information management issues that inevitably surface during these analyses.

i Abstract Davis, Raymond J.; Dugger, Katie M.; Mohoric, Shawne; Evers, Louisa; Aney, William C. 2011. Northwest Forest Plan—the first 15 years (1994–2008): status and trends of northern spotted owl populations and habitats. Gen. Tech. Rep. PNW- GTR-850. Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station. 147 p. This is the second in a series of periodic monitoring reports on northern spotted owl (Strix occidentalis caurina) population and habitat trends on federally administered lands since implementation of the Northwest Forest Plan in 1994. Here we summarize results from a population analysis that included data from long- term demographic studies during 1985–2008. This data was analyzed separately by study area, and also in a meta-analysis across all study areas to assess temporal and spatial patterns in fecundity, apparent survival, recruitment, and annual rates of population change. Estimated rates of annual population decline ranged from 0.4 to 7.1 percent across federal study areas (weighted average of 2.8 percent). Covariates for barred owls (Strix varia), weather, climate, habitat, and reproductive success were analyzed and had varying degrees of association with owl demographic parameters. We now have more evidence that increasing numbers of barred owls and loss of nesting/roosting habitat contributed to demographic declines in some study areas. We also summarize results from a habitat analysis that used the above data in conjunction with remotely sensed data from 1994 to 2007 to develop “habitat suitability” models and habitat maps. These maps were used to quantify the amount and distribution of owl habitats. We also report on causes of habitat change during this period. On federal lands, nesting/roosting habitat declined by 3.4 percent rangewide, with some physiographic provinces experiencing losses of 10 percent. Dispersal habitat increased by 5.2 percent, but dispersal-capable landscapes declined by 1 percent. Wildfire remains the leading cause of habitat loss. We developed a rangewide “wildfire suitability” model and map to illuminate the portions of the owl’s range where suitable nesting/roosting habitat overlaps with landscapes suitable for the occurrence of large wildfires. Barred owls and management of owl habitat in fire-prone areas continue to be topics for future monitoring, research, and management consideration. Keywords: Northwest Forest Plan, effectiveness monitoring, northern spotted owl, geographic information system, owl habitat, habitat suitability, wildfire suitability, demographic study, remote sensing, predictive model, habitat model.

ii Summary For the eight federal study areas associated with the Northwest Forest Plan (the Plan) effectiveness monitoring program, the average rate of population decline was 2.8 percent per year. Strong evidence of declines in annual rates of population change were reported for five of the eight individual effectiveness monitoring area study sites, but confidence limits on point estimates for three areas in the center of the northern spotted owl (Strix occidentalis caurina) range (southwest Oregon) overlapped lambda = 1.0, suggesting these three populations may not be declining. Rates of population decline were highest in the northern portions of the owl’s range (Washington and northern Oregon) where populations are estimated to have declined 40 to 60 percent since the Plan’s implementation. A variety of covariates including presence of barred owls (Strix varia), weather and long-term climate cycles, the amount of suitable nesting/roosting habitat on and adjacent to each study area, and the previous year’s reproductive success, were included in the analysis of demographic data to explore associations between them and observed population trends. These covariates had varying degrees of association with owl demographic parameters, but at least one vital rate (i.e., fecundity, apparent survival, or population) was declining on all study areas. The long-term demographic data we continue to collect are the key to understanding the range of factors that are affecting the recovery of spotted owl populations. At present, the invasion of the competitive barred owl and the amount of suitable nesting/roosting habitat are the factors most associated with spotted owl vital rates. Directly managing barred owl encroachment into spotted owl habitats may be beyond the scope of the Plan, but maintaining large blocks of suitable spotted owl habitat will likely play a key role in decreasing negative interactions between the two species and increasing the likelihood of the persistence of spotted owl populations. On federal lands, we estimated nesting/roosting habitat losses for 1994 through 2007 in California, and 1996 through 2006 in Oregon and Washington at 3.4 percent rangewide. Although rangewide losses have not yet exceeded what was anticipated under the Plan, some physiographic provinces have incurred losses up to 10 percent. This and the fact that most of the nesting/roosting habitat loss occurred within reserved land use allocations, and not within the federal matrix outside of these reserves, raises some concern. But in spite of this paradox, the large, repetitive design of reserves appears to still be functioning as intended. Of the 12 million ac of nesting/roosting habitat remaining, 71 percent occurs on federally administered lands, and approximately 70 percent of this is in reserved land use allocations (not including riparian reserves). Over half of the nesting/roosting habitat occurs in the central (core) portions of the owl’s range, within the Klamath Mountain provinces of Oregon and California (27 percent) and the western Cascades of Oregon (26 percent). Not enough time has yet elapsed for us to accurately detect or estimate any significant recruitment of nesting/roosting habitat; however, increases were observed in “marginal” (younger) forests indicating that future recruitment of nesting/roosting habitat is on track to occur, as anticipated, within the next few decades.

iii In addition to providing potential future nesting/roosting habitat, some younger forests function as dispersal habitat. Forest succession accounted for some dispersal habitat recruitment, especially in the more productive tree-growing portions of the range (i.e., Oregon Coast Range). Partial disturbances of nesting/roosting habitat also accounted for some of this recruitment as well. Loss of dispersal habitat, primarily from wildfires, was observed, but recruitment rates exceeded losses, resulting in a net increase in dispersal habitat of 5.2 percent (rangewide). In spite of this net gain, dispersal-capable landscapes actually decreased by 1 percent within the owl’s range because of the spatial distribution of this habitat. Even with this small decrease, the network of large reserves remains fairly well connected, with the exception of the northern portion of the eastern Cascades of Wash- ington and also within the southern tip of the range where some large reserves appear to be isolated (including the Marin County population). Recent improvements in remotely sensed vegetation and change-detection mapping has resulted in better habitat maps to replace the baseline versions produced for the first monitoring report. Progress in habitat “niche” modeling methods and software has improved our ability to map not only habitat for spotted owls, but also “suitable habitat” for large wildfires. Wildfire remains the leading cause of owl habitat loss. About 3.6 million ac of nesting/roosting habitat remain in landscapes that are naturally prone to large wildfires. Most of this “fire-prone” habitat (85 percent) occurs within the “core” of the owl’s range (i.e., the Klamath Mountains and the western Cascades of Oregon). Not all habitat burned is lost to owls, as fire intensity and frequency play a role in the effect of fire on owl habitat use. Our monitoring showed that large wildfires resulted in 30 to 62 percent loss of the nesting/roosting owl habitat within their perimeters. Wildfire is a natural ecological process under which northern spotted owls have evolved, but the landscapes in which this occurred were heavily altered during the 20th century. Most remaining nesting/roosting habitat is now contained on federal land, and its fragmented condition makes it, and the populations that rely on it, more vulnerable to future large wildfires. Conservation management for northern spotted owls in relation to wildfire will involve understanding (1) where suitable owl habitats overlap suitable habitat for large wildfire; (2) the effect of fuel reduction treatments to reduce fire risk on owl habitat use and demographics; and (3) the relationships of fire frequency, severity, and extent with owl habitat use and demographics.

iv Contents 1 Chapter 1: Introduction and Background Raymond J. Davis, Katie M. Dugger, and Shawne Mohoric 3 References 5 Chapter 2: Population Status and Trend Katie M. Dugger and Raymond J. Davis 5 Introduction 5 Data Sources and Methods 7 Field Data Collection 7 Data Analysis 7 Error Checking 9 Estimating Survival 9 Estimating Fecundity 0 1 Estimating Annual Rate of Population Change and Realized Population Change 01 Results 1 1 Survival 2 1 Fecundity 4 1 Annual Rate of Population Change 5 1 Realized Population Change 15 Discussion 17 Summary 18 References 21 Chapter 3: Habitat Status and Trend Raymond J. Davis and Katie M. Dugger 21 Introduction 32 Habitat Monitoring Under the Plan 24 Methods and Data Sources 4 2 Land Use Allocation Data 7 2 Vegetation Data 8 2 Change-Detection Data 0 3 Spotted Owl Presence Data 0 3 Habitats, the Niche Concept, and Habitat Modeling 2 3 Habitat Modeling Process 3 3 Environmental Variables 4 3 Modeling Regions 6 3 Habitat Map Development and Evaluation 8 3 Nesting/Roosting Habitat 0 4 Dispersal Habitat

v 1 4 Habitat Fragmentation 43 Results 3 4 Habitat Suitability Modeling 3 4 Nesting/Roosting Habitat 94 Dispersal Habitat 2 5 Habitat Fragmentation 53 Discussion 54 Summary 55 References 63 Chapter 4: Large Wildfires Within the Owl’s Range Raymond J. Davis, William C. Aney, Louisa Evers, and Katie M. Dugger 63 Introduction 65 Methods and Data Sources 67 Environmental Data 9 6 Large Wildfire Data 0 7 Wildfire Suitability Modeling 71 Results 73 Discussion 78 Summary 08 References 87 Chapter 5: Emerging Issues, Related Research, and Research Needs Katie M. Dugger and Raymond J. Davis 87 Emerging Issues 89 Related Research and Research Needs 29 Summary 29 References 96 Acknowledgments 97 Metric Equivalents 99 Appendix A: Environmental Variables Used for Habitat Suitability Modeling 103 Appendix B: Nearest Neighbor Distance Analysis of Demographic Study Area Data 107 Appendix C: Habitat Suitability Modeling Replicate Data 121 Appendix D: Nesting/Roosting Habitat Status and Trend Tables Based on LandTrendr Analysis 125 Appendix E: Dispersal Habitat Status and Trend Tables Based on LandTrendr Analysis 129 Appendix F: Crosswalk for Modifying Bookend 2 (2006/07) Map for Making Habitat Suitability Histograms 135 Appendix G: Wildfire Suitability Modeling, MaxEnt Replicate Data 141 Appendix H: Regional Inventory Plot Analysis vi Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Chapter 1: Introduction and Background Raymond J. Davis, Katie M. Dugger, and Shawne Mohoric

In 1994, the Northwest Forest Plan (referred to hereafter as The first monitoring effort reporting on status and the Plan) amended 19 existing Forest Service and 7 Bureau trends of northern spotted owl populations and habitat (Lint of Land Management resource management plans within 2005) included a summary of the fourth northern spotted the range of the northern spotted owl (Strix occidentalis owl meta-analysis (Anthony et al. 2006) and produced a caurina). An interagency effectiveness monitoring frame- habitat baseline map using the latest technology and best work was implemented to meet requirements for tracking available data at the time. This report covers the first 15 the status and trends for late-successional and old-growth years of implementation under the Plan, including a summa- forests, northern spotted owl populations and habitat, ry of the fifth northern spotted owl population meta-analysis marbled murrelet (Brachyramphus marmoratus) popula- (Forsman et al. 2011) and the development of new habitat tions and habitat, watershed condition, social and economic maps based on new vegetation data, analytical methods, and conditions, and tribal relationships. Monitoring results habitat modeling technologies. are reported at 1-year intervals and evaluated at 5-year Lint (2005) realized that as technology advances, there intervals. The first regional monitoring reports roughly will be a need to refine or adapt old monitoring methods for covered the first 10 years of Plan implementation and were new analytical approaches. With the help of leaders in the documented in a series of General Technical Reports posted fields of statistics and wildlife demographics, the analyti- at http://www.fs.fed.us/pnw/publications/gtrs.shtml. The cal methods for conducting the population meta-analysis first northern spotted owl population and habitat monitoring continue to advance. Barred owl (Strix varia), climate, and report was produced in 2005 covering status and trends of habitat covariates were included in the latest analysis for the populations up to 2003 and habitat up to 2002 (Lint 2005). first time in 2009 (Forsman et al. 2011). The habitat covari- This report is the second in the series of northern spotted ates used were products from the 10-year report (Davis and owl effectiveness monitoring reports (Lint et al. 1999) and Lint 2005). The inclusion of these new modeling techniques covers population status and trend up to 2008 and habitat and covariates allowed us to investigate relationships be- status and trend up to 2007. tween them and owl demographics for the very first time. The goal of the northern spotted owl monitoring Likewise, the habitat analysis has evolved to incor- program is to evaluate the success of the Plan in arresting porate new habitat modeling and forest pattern analysis the downward trends in populations and habitats that were software that can be used for identifying habitat conditions, largely responsible for the establishment of the Plan. In part, characterization of change to those conditions, and the the Plan was designed to maintain and restore habitat condi- recruitment of those conditions through forest succession. tions necessary to support viable populations of the north- Improvements were made to the vegetation data used to ern spotted owl on federally administered lands throughout characterize owl habitat, including the addition of more the owl’s range (fig. 1-1). The objectives for northern spotted variables for habitat modeling and analysis. Most notable, owl effectiveness monitoring are as follows: a consistent vegetation data set was produced for the entire range of the northern spotted owl, which has never been 1. Assess changes in population trends and demographic available before. This new vegetation data set replaces the rates of spotted owls on federal lands within the owl’s two previously used data sets (IVMP and CALVEG) and, range. along with new modeling software, allowed us to refine 2. Assess changes in the amount and distribution of the previous baseline habitat map. Therefore the baseline nesting, roosting, foraging and dispersal habitat for amounts and distribution of owl habitat reported in the 10- spotted owls on federal lands. year report are replaced by results presented in this report.

1 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 1-1—The range of the northern spotted owl. NASA = National Aeronautics Space Administration.

2 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Improvements were also made to the remotely sensed Implementation of phase II depends on our ability to relate data used for estimating habitat changes. These improve- owl demography to habitat conditions such that we can ments include a finer time sequence of change-detection relate habitat status and trends directly to population status (annual versus 4- to 5-year intervals) and an improved and trends with acceptable confidence. To date, attempts to ability to detect lower intensity disturbances (i.e., thinning, develop predictive models have had mixed results (Dugger insects, and disease). Another improvement in our ability to et al. 2005, Franklin et al. 2000, Olson et al. 2004) and have detect habitat changes came from the creation of a vegeta- generally been unsuccessful across the range of the owl; tion data set that contains the same variables as the baseline however, some progress has been made as noted above and, data set, but for a later period. We called these vegetation as technology continues to advance, this remains our goal. data sets “bookends.” Our first bookend is from 1994 in After 15 years, agency managers continue to be California and from 1996 in Oregon and Washington. The proactive and supportive of the monitoring program. As other bookend is from 2007 in California and from 2006 in Lint (2005) stated, this support is, “of utmost importance to Oregon and Washington. Therefore our habitat maps and the future of the effectiveness monitoring program.” The our analysis of habitat status and trends cover the period Northwest Forest Plan’s effectiveness monitoring program from 1994/96 to 2006/07. (Mulder et al. 1999) has received national and international The spotted owl monitoring plan includes two phases attention (Gosselin 2009) and has been noted as the largest of monitoring (Lint et al. 1999). Phase I entails demographic and most comprehensive regional forest plan monitoring monitoring of individual territorial owls on eight federal ever conducted (McAlpine et al. 2007). The monitoring data study areas to estimate population demographics including created and the analysis results presented in the 10-year survival, fecundity, and rate of population change while monitoring report have provided valuable information for also tracking habitat conditions rangewide. The eight managers and policymakers in making informed decisions. federal study areas that are part of phase I occur on federal Examples include northern spotted owl recovery planning lands administered by the U.S. Forest Service, Bureau of (USDI 2008b) and designation of critical habitat (USDI Land Management, and the National Park Service. They 2008a) and increased emphasis by regulatory and manage- provide population trend data for a representative mix of ment agencies to reduce risk of owl habitat and old forests areas considered key to the success of owl management from high-severity fire in dry provinces (Spies et al. 2006). under the Plan. The scientists who developed the monitoring plan determined that these eight study areas were the References minimum number needed to be able to make scientifically Anthony, R.G.; Forsman, E.D.; Franklin, A.B.; credible and defensible inferences of population trends to Anderson, D.R.; Burnham, K.P.; White, G.C.; the broader federal landscape within the owl’s range (Lint Schwarz, C.J.; Nichols, J.D.; Hines, J.E.; Olson, et al. 1999, Mulder 1997). It is hoped that eventually the G.S.; Ackers, S.H.; Andrews, L.W.; Biswell, B.L.; demographic monitoring data can be combined with the Carlson, P.C.; Diller, L.V.; Dugger, K.M.; Fehring, habitat monitoring data to develop predictive models of K.E.; Fleming, T.L.; Gearhardt, R.P.; Gremel, owl occurrence and demographic performance based on S.A.; Gutiérrez, R.J.; Happe, P.J.; Herter, D.R.; observed habitat conditions. This would allow for imple- Higley, J.M.; Horn, R.B.; Irwin, L.L.; Loschl, P.J.; mentation of phase II, which increases emphasis on habitat Reid, J.A.; Sovern, S.G. 2006. Status and trends in monitoring and decreases the population monitoring to a demography of northern spotted owls, 1985–2003. minimum of four study areas, which would provide a means Wildlife Monographs No. 163. 48 p. to validate the population predictions of the habitat models.

3 GENERAL TECHNICAL REPORT PNW-GTR-850

Davis, R.; Lint, J. 2005. Habitat status and trend. In: PNW-GTR-648. Portland, OR: U.S. Department of Lint, J., tech. coord. Northwest Forest Plan—the first Agriculture, Forest Service, Pacific Northwest Research 10 years (1994–2003): status and trends of northern Station. 176 p. spotted owl populations and habitat. Gen. Tech. Rep. McAlpine, C.A.; Spies, T.A.; Norman, P.; Peterson, A. PNW-GTR-648. Portland, OR: U.S. Department of 2007. Conserving forest biodiversity across multiple land Agriculture, Forest Service, Pacific Northwest Research ownerships: lessons from the Northwest Forest Plan and Station: 21–82. the Southeast Queensland regional forests agreement Dugger, K.M.; Wagner, F.; Anthony, R.G.; Olson, G.S. (Australia). Biological Conservation. 134: 580–592. 2005. The relationship between habitat characteristics Mulder, B. 1997. Supplemental information for the and demographic performance of northern spotted owls policy and management review: general monitoring in southern Oregon. Condor. 107: 863–878. program, late-succession old-growth, northern spotted Forsman, E.D.; Anthony, R.G.; Dugger, K.M.; Glenn, owl and marbled murrelet. Report by the Effectiveness E.M.; Franklin, A.B.; White, G.C.; Schwarz, C.J.; Monitoring Team. Unpublished report. On file with: U.S. Burnham, K.P.; Anderson, D.R.; Nichols, J.D.; Hines, Department of Agriculture, Forest Service, 333 SW 1st J.E.; Lint, J.B.; Davis, R.J.; Ackers, S.H.; Andrews, Avenue, Portland, OR 97208-3623. L.S.; Biswell, B.L.; Carlson, P.C.; Diller, L.V.; Mulder, B.S.; Noon, B.R.; Spies, T.A.; Raphael, M.G.; Gremel, S.A.; Herter, D.R.; Higley, J.M.; Horn, R.B.; Palmer, C.J.; Olsen, A.R.; Reeves, G.H.; Welsh, Reid, J.A.; Rockweit, J.; Schaberl, J.; Snetsinger, T.J.; H.H., tech. coords. 1999. The strategy and design of Sovern, S.G. 2011. Population demography of northern the effectiveness monitoring program for the Northwest spotted owls. Studies in Avian Biology 40. Berkeley, CA: Forest Plan. Gen. Tech. Rep. PNW-GTR-437. Portland, University of California Press. 106 p. OR: U.S. Department of Agriculture, Forest Service, Franklin, A.B.; Anderson, D.R.; Gutiérrez, R.J.; Pacific Northwest Research Station. 138 p. Burnham, K.P. 2000. Climate, habitat quality, and fit- Olson, G.S.; Glenn, E.M.; Anthony, R.G.; Forsman, ness in northern spotted owl populations in northwestern E.D.; Reid, J.A.; Loschl, P.J.; Ripple, W.J. 2004. California. Ecological Monographs. 70(4): 539–590. Modeling demographic performance of northern spotted Gosselin, F. 2009. Management on the basis of the best owls relative to forest habitat in Oregon. Journal of scientific data or integration of ecological research Wildlife Management. 68: 1039–1053. within management? Lessons learned from the northern Spies, T.A.; Hemstrom, M.A.; Youngblood, A.; Hummel, spotted owl saga on the connection between research and S. 2006. Conserving old-growth forest diversity in management in conservation biology. Biodiversity and disturbance-prone landscapes. Conservation Biology. 20: Conservation. 18: 777–793. 351–362. Lint, J.; Noon, B.; Anthony, R.; Forsman, E.; Raphael, U.S. Department of the Interior [USDI]. 2008a. M.; Collopy, M.; Starkey, E. 1999. Northern spotted Endangered and threatened wildlife and plants; revised owl effectiveness monitoring plan for the Northwest designation of critical habitat for the northern spotted Forest Plan. Gen. Tech. Rep. PNW-GTR-440. Portland, owl; final rule. Federal Register. 73(157): 47326–47522. OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station. 43 p. U.S. Department of the Interior [USDI]. 2008b. Final recovery plan for the northern spotted owl Lint, J.B., tech. coord. 2005. Northwest Forest Plan—the (Strix occidentalis caurina). 142 p. On file with: U.S. first 10 years (1994–2003): status and trends of northern Department of the Interior, Fish and Wildlife Service, spotted owl populations and habitat. Gen. Tech. Rep. 2600 SE 98th Ave., Suite 100, Portland, OR 97266.

4 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Chapter 2: Population Status and Trend Katie M. Dugger and Raymond J. Davis

Introduction et al. (2011). The objectives of the most recent population The collection of demography data is the foundation of the status and trend meta-analysis were as follows: effectiveness monitoring program for northern spotted owls • Estimate age-specific survival and fecundity rates (Strix occidentalis caurina) (Lint et al. 1999), designed to and their sampling variances for individual study monitor the effect of the Northwest Forest Plan (the Plan) on areas. populations. Demographic surveys for spotted owls follow- • Determine if any trends in adult female survival and ing standardized data collection protocols began on some fecundity exist across study areas. study areas as early as 1985 (northwest California: Franklin • Estimate annual rates of population change (λ) and et al. 1996a, 1996b) even before the monitoring plan was their sampling variances for individual study areas. actually finalized. The first rangewide meta-analysis was • Determine if the declines in apparent survival and conducted in 1991 (Anderson and Burnham 1992), then populations, which were documented previously again in 1993 (Burnham et al. 1996), and every 5 years (Anthony et al. 2006), have continued or stabilized. thereafter (1998: Franklin et al. 1999; 2004: Anthony et • Determine whether changes in the amount of suit- al. 2006; 2009: Forsman et al. 2011). This long history of able habitat, the presence of barred owls (Strix owl surveys and demographic data collection represents varia), or climate explain the observed annual vari- the single largest, long-term mark-recapture data set in the ability in owl vital rates. world for a threatened species (Courtney et al. 2004), and • Estimate components of the rate of population these data are invaluable for monitoring spotted owls under change, including apparent survival and recruit- the Plan. ment rates that were not done in previous analyses The goal of the population component of the monitoring (Anthony et al. 2006, Burnham et al. 1996, Franklin program is to determine if the Plan is arresting or slowing et al. 1999). the declining trend in northern spotted owl populations Data Sources and Methods on federally administered lands throughout the owl’s Data from eight demographic study areas in Washington, range. This is accomplished with annual data collection on Oregon, and California were used to estimate status eight federal study areas associated with the effectiveness and trends of owl populations on federal lands (fig. 2-1). monitoring plan (Lint et al. 1999). For the 10-year report Although it is not part of the monitoring plan, data from (Lint 2005), these eight areas and data from three other the Rainier study area in Washington were also included be- independent study areas provided relevant data to address cause the study area occurs primarily on federal land. The this question on federal lands managed under the Plan (An- two additional study areas in the latest meta-analysis are thony et al. 2006). After 15 years, we report results from the the Hoopa on tribal lands and the Green Diamond Resource eight federal demographic study areas and one independent study area on private timber company lands (Forsman et al. study area. These nine areas are spread throughout the owl’s 2011). Because Hoopa and Green Diamond Resources did range (fig. 2-1) and data on owl occupancy, survival, and not include any lands managed under the Plan, they were productivity were gathered annually from each to estimate excluded from this monitoring report, except when meta- apparent adult survival, reproduction, and annual rate of analysis results including all 11 study areas are presented. change of owl populations. Detailed results of the analyses This monitoring report is based on nine study areas of these data and data from two other, independent study managed under the Plan that include variation in climate, areas within the range of the owl are reported by Forsman vegetation, and topography and encompass most of the

5 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 2-1—Location of nine demography study areas comprising primarily federal lands administered under the Northwest Forest Plan and included in the 2009 northern spotted owl meta-analysis. Source: Forsman et al. (2011).

6 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

northern spotted owl’s geographic distribution. The forests Corvallis, Oregon, to analyze the data from 11 study areas. on all study areas are dominated by conifers or mixtures This workshop was led by research scientists with inter- of conifers and hardwoods, although there are regional nationally recognized expertise in population dynamics, differences in species composition (for more details, see statistics, and the analysis of capture-recapture data. The Forsman et al. 2011). The nine study areas range from 396 analyses were conducted under the direct guidance of these to 1514 mi2; the median study area size was 691 mi2, and scientists. Consistent with the previous four workshops con- the mean was 829 mi2 (table 2-1). These nine study areas vened since 1991 to analyze spotted owl demographic data, encompassed 7460 mi2 or approximately 8 percent of the all participants adopted formal protocols for error-checking owl’s range, and the numbers of years included in these data data sets and for the development of a priori model sets for sets ranged from 17 (Rainier) to 24 (Northwest California). each parameter of interest (Anderson et al. 1999). Thus, the Four of these study areas (Olympic, H.J. Andrews, South data were collected and prepared in a consistent manner Cascades, and Northwest California) primarily comprised among study areas, and there were no analyses of additional federal lands administered by USDA Forest Service, the models after post hoc examination of initial results (i.e., USDI Bureau of Land Management, and the USDI National all data sets were analyzed the same way). Detailed results Park Service (table 2-1). The other five (Cle Elum, Rainier, from this workshop (summary presented here) are reported Coast Ranges, Tyee, and Klamath) included a mixture of in Forsman et al. (2011), and these analyses represent a federal, private, and state lands intermixed in a checker- retrospective, observational study, which assesses the board pattern of ownership (table 2-1). strength of association between owl vital rates and a variety of explanatory covariates rather than addressing direct Field Data Collection cause-effect relationships. Data on individually identifiable (i.e., banded) owls were collected from the nine demographic study areas annually. Error Checking During each breeding season (March through August), Crew leaders from each study area compiled survival, multiple visits (usually > three per season) were made to fecundity, and rate of population change data sets in a owl territories to locate banded owls; confirm band num- consistent manner, following specific instructions provided bers, sex, and age; and band any unmarked owls. In addi- by workshop organizers. When digital files were completed, tion, the number of young produced was documented for data entry was error checked by independent members of each territorial owl, and fledglings were banded resulting the workshop organizing team. The capture-history files in a known-age population of spotted owls on each study for estimation of survival and annual rate of population area. For details on the standardized field methods used change were error checked by randomly drawing 10 capture to capture, mark, age, sex, and estimate productivity, see histories from each study area file and comparing them to Franklin et al. (1996a). These methods resulted in complete paper copies of the field data that supported each of these capture histories over time of every owl banded during this capture histories. Fecundity data entry was error checked in study and the number of young fledged per territorial female a similar way, with 10 records of reproductive success for a (NYF) located each year. From these data, annual apparent specified female in a given year compared to paper copies of survival (φ) by sex and age, annual productivity (NYF) the field data forms. If errors were found in the first round by age, and the annual rate of population change (λ) were of checking, the errors were corrected and the process was estimated (Forsman et al. 2011). repeated with another sample of 10 records. If errors were found in the second round of data checking, the entire file Data Analysis was returned to the crew leader and principal investigator th th During a 9-day period in January 2009 (9 through 17 ), for review and correction. This sequence of error checking a workshop was held at Oregon State University in and correction was continued until no errors were found

7 GENERAL TECHNICAL REPORT PNW-GTR-850 b

otal 583 T 2,550 2,364 3,306 2,315 1,510 3,082 2,800 1,170 19,680 encounters

1 21 474 602 649 490 388 576 650 153

Total Total 4,193

280 479 486 243 337 457 347 148 133 2,910 Adults

of banded owls S2 80 80 97 10 32 91 32 12 1 668 134

a Number 8 14 43 66 19 28 31 S1 1 615 137 169

396 861 619 549 689 837 691 mi size Study 7,460 1,304 1,514

egion

Ecological r ashington Douglas-fir ashington mixed conifer ashington Douglas-fir Oregon Cascades Douglas-fir Oregon coastal Douglas-fir Oregon coastal Douglas-fir W Oregon Cascades Douglas-fir Oregon/California mixed conifer W W Oregon/California mixed conifer

class

Federal Mixed Mixed Federal Federal Mixed Mixed Mixed Federal Land- owner

ears Y

1991–2008 1990–2008 1990–2008 1990–2008 1988–2008 1990–2008 1989–2008 1992–2008 1985–2008

ovince pr estern Cascades Physiographic W Coast Ranges Coast Range Olympic Peninsula W Klamath Eastern Cascades W Eastern Cascades Klamath estern and c

estern Cascades c c c

c c

c T California Age class codes indicate age at which owls were banded and became part the of mark-recapture data = 1 year set: old, = 2 years S1 S2 old, and adults ≥ 3 years old. All captures, recaptures, and resightings, excluding multiple encounters individuals of in the same year. One eight of study areas monitored under the northern spotted effectiveness owl monitoring program for the Northwest Forest Plan. Source: adapted from Forsman et al. (2011). Table 2-1−Descriptions nine of Table demographic study areas associated with landPlan managed Forest under Northwest the Study area Washington: Cle Elum Rainier Olympic Oregon: Coast Ranges Andrews H.J. Tyee Klamath South Cascades California: Northwest a b otals c

8 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

in 10 randomly drawn records, although it is possible that and climate covariates, were also included in the analysis. a low level of data entry error might still persist. Copies The nature (positive or negative) of these effects was of error-checked records and field data forms submitted to hypothesized a priori, and the appropriate models reflecting confirm these records were archived, and all crew lead- these effects were included in the initial model sets prior to ers signed statements before submitting data for analysis analysis. certifying the accuracy of their data. The meta-analysis of all 11 study areas combined was conducted in a similar fashion, but in addition to study area, Estimating Survival time trends, the cost of reproduction, and the barred owl Cormack-Jolly-Seber open population models (CJS) covariate, models also included land ownership, ecological (Franklin et al. 1996a, Lebreton et al. 1992) in Program region, latitude, climate, and habitat change. MARK (White and Burnham 1999) were used to estimate apparent survival of owls each year. Because survival esti- Estimating Fecundity mates from CJS models cannot separate losses of individu- All analyses of reproductive rate were based on the annual als who died from losses owing to permanent emigration, number of young produced per territorial female (NYF), but these models estimate apparent survival, which incorporates to be consistent with previous reports (Anthony et al. 2006, the annual site fidelity of individuals (true survival x site Forsman et al. 1996, Franklin et al. 1999), estimates from fidelity = apparent survival). Spotted owls show high annual these models were presented as “fecundity,” where fecundi- site fidelity (Forsman et al. 2002), so permanent emigration ty is the average annual number of female young produced does not seriously bias model estimates, and apparent sur- per female owl (NYF/2). This adjustment assumes a 1:1 sex vival is believed to be very close to true survival (Anthony ratio at birth, which has been supported by previous genetic et al. 2006, Forsman et al. 2011). The general approach used analyses of blood collected from juveniles (Fleming et al. to generate survival estimates from capture-recapture data 1996). Models were developed a priori to investigate the ef- on individual study areas was as follows: fects of age, general time variation, a variety of time trends, the proportion of owl territories where barred owls were • Decide on a set of a priori models for analysis and detected each year, and an even-odd year effect, which has the order in which models will be run. previously been shown to reflect a temporal cycle in spotted • Evaluate goodness-of-fit of the data to the general owl reproduction (i.e., Anthony et al. 2006). In addition, CJS model and estimate an over-dispersion param- climate and habitat covariates were included in the analysis. eter (c-hat = ĉ) using the median c-hat approach in The general approach used to generate fecundity estimates Program MARK. was as follows: • Use the estimated c-hat to adjust covariance matri- ces for over-dispersion and to obtain quasi-Akaike’s • Decide on a set of a priori models for analysis and

information criteria (QAICc) for model selection. the order in which models will be run. • Run all models for capture probability and apparent • Determine whether spatial variance (the random ef- survival developed in the pre-analysis a priori model fect of territory) should be included in the modeling set. process. • Select appropriate models for inference based • Use Proc Mixed in SAS (SAS Institute Inc. 2008) to

on QAICc model selection results (Burnham and fit all a priori models to the annual averages of NYF Anderson 2002). using a regression model based on a normal distribution. Several covariates expected to affect survival, includ- • Select appropriate models for inference based ing age, sex, the cost of reproduction, the proportion of on QAIC model selection results (Burnham and territories where barred owls were detected each year, c Anderson 2002).

9 GENERAL TECHNICAL REPORT PNW-GTR-850

There was no consistent pattern regarding the best In addition to an analysis of annual population change model for fecundity among study areas, so a nonparametric for each individual study area, a meta-analysis was conduct- approach was used to estimate mean NYF by age class. The ed with all 11 study areas combined, where landownership, mean NYF was computed for each year and age class. Then latitude, climate and weather, and ecological region were these means were averaged across years within each age also included. Analysis details and meta-analysis results are class. The estimated standard error was computed as the reported in Forsman et al. (2011). standard error of the average of the averages among years. Estimates of realized population change (D ) were also This method gave equal weight to all years, regardless of t computed and reflect the proportional change in estimated the number of birds actually observed, and it did not force a population size relative to population size in the initial year model for changes over time. of analysis, and were computed following the methods of As was done for survival, a meta-analysis of fecundity Franklin et al. (2004). On each study area, annual estimate with all 11 study areas combined was conducted, and in of realized population change was calculated as: addition to the covariates included in the individual study area analysis, land ownership, latitude, climate, and ecological region were also included. Analysis details and meta- analysis results are reported in Forsman et al. (2011).

Estimating Annual Rate of Population Change and where x was the year of the first estimated λt. For example, Realized Population Change given three, year-specific lambdas of say 0.9 in 1993, 1.2 in The reparameterized Jolly-Seber method (Pradel 1996) was 1994, and 0.7 in 1995, the realized population change would be 0.9 x 1.2 x 0.7 = 0.756. This value means that at the end used to estimate annual rates of population change (λRJS) in Program MARK using capture-recapture data. A param- of 1995, the population was 75.6 percent of the starting population in 1993. Thus, estimates of realized population eterization was used to generate annual estimates of λ (λt) for each study area, which allowed for decomposition of λ change clearly illustrate the long-term, cumulative trends in into two components, apparent survival (φ) and recruitment annual population changes. (f), where: Results λ = φ +f t t t The following is a summary of the demographic analysis of apparent survival, fecundity, annual rate of population Apparent survival (φt) reflects both survival of territory change, and realized population change for the northern holders within study areas and site fidelity at time t (year), spotted owl reported by Forsman et al. (2011). These so both death and permanent emigration are included in this analyses are the most long-term and comprehensive to date parameter. Recruitment (ft) is the number of new owls in across the range of the owl; however, although the 11 study the population at time t+1 per animal in the population at areas included in this analysis covered a large portion of the time t and reflects both individuals born on the study area owl’s geographic range, they were not randomly selected. that become established territory holders, and immigration Thus, results cannot be considered representative of owl of recruits from outside the study area. Thus, the estimate populations throughout its entire range and cannot be used of λt accounts for all of the losses and gains in the study to assess demographic trends on nonfederal lands because area populations during each year and results in minimum only two study areas on nonfederal lands were included bias in estimation of the annual rate of population change in the analysis. However, Forsman et al. (2011) believed (Anthony et al. 2006). their results to be representative of most owl populations

10 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

on federal lands as they include nine large study areas, rate (fledglings per pair) in the previous year, and weather with comprehensive geographic coverage and a variety of had varying effects on survival depending on the study landownership and management strategies. Thus, the results area. Mean annual estimates of model-averaged apparent from the nine study areas associated with federal land survival of female owls ranged from 0.529 to 0.794 for managed under the Plan can be used to make inferences to 1-year-olds, 0.674 to 0.864 for 2-year-olds, and 0.819 to populations on those lands. 0.865 for adults (≥3 years old) (table 2-2). Most notably, survival was declining on all but the Klamath study area, and Survival in some cases, the declines occurred primarily in the last For the nine individual study areas, the number of banded 10 years or so (Coast Ranges, H.J. Andrews, Tyee, South owls included in the survival analysis were 615 1-year-olds, Cascades). Declines were most evident in Washington and 668 2-year-olds, and 2,910 adults (>3 years old) with 19,680 strongest in the last 5 years for the Cle Elum and Rainier total encounters across all individuals and age classes (table study areas. The Klamath study area was the only one for 2-1). The number of recaptures in this data set was 4.5 times which no trend in survival was observed, although large the number of initial captures. amounts of annual variation in adult survival were observed In general, survival was similar between sexes (except (see fig. 5b in Forsman et al. 2011). for Olympic where survival was higher for males) and For the Rainier and Olympic study areas in Washing- higher for adults compared to subadults (table 2-2). Factors ton, survival was negatively associated with high rates of including time and time trends, the proportion of territories reproduction in the previous year, but this effect was not where barred owls were detected each year, reproductive evident on any of the other study areas. In the meta-analysis

Table 2-2—Average survival rates with standard errors (SE) for female northern spotted owls by age class in the nine demographic study areas associated with land managed under the Northwest Forest Plan Age class Landowner 1 year old 2 years old ≥3 years old Study area class Survivala SE Survivala SE Survivala SE Washington: Cle Elumb Mixed 0.794 0.051 0.820 0.023 0.819 0.013 Rainier Mixed 0.541 0.181 0.674 0.156 0.841 0.019 Olympicb Federal 0.529 0.148 0.786 0.081 0.828 0.016 Oregon: Coast Rangesb Mixed 0.742 0.072 0.864 0.031 0.859 0.009 H.J. Andrewsb Federal 0.717 0.084 0.830 0.042 0.865 0.010 Tyeeb Mixed 0.761 0.043 0.864 0.020 0.856 0.008 Klamathb Mixed 0.788 0.040 0.858 0.020 0.848 0.008 South Cascadesb Federal 0.692 0.069 0.733 0.053 0.851 0.010 California: Northwest Californiab Mixed 0.774 0.031 0.784 0.031 0.844 0.009 Note: See table 2-1 for data years. a Average survival is the arithmetic mean of model-averaged annual survival estimates for females. Standard errors were calculated using the delta method. b One of eight study areas monitored under the northern spotted owl effectiveness monitoring program for the Northwest Forest Plan. Source: adapted from Forsman et al. (2011).

11 GENERAL TECHNICAL REPORT PNW-GTR-850

of all 11 study areas, the negative cost of reproduction et al. 2011). Mean fecundity was highest for adults (0.330, on survival was an important covariate and a consistent SE = 0.025), lower for 2-year-olds (0.202, SE = 0.042), and effect across all study areas. The analyses of individual nearly negligible (0.07, SE = 0.015) for 1-year-olds (Forsman study areas supported the negative effect of barred owls et al. 2011). on survival, but the effect was variable among study areas: Fecundity differed greatly by study area, and adult decreased survival was associated with higher proportions fecundity was highest on Cle Elum (0.553, SE = 0.052) and of territories where barred owls were detected for Rainier, lowest in the Coast Ranges (0.263, SE = 0.04) (table 2-3). Coast Ranges, and H.J. Andrews, with weaker evidence There was considerable annual variation in fecundity, but found for the Olympic and Northwest California, and neg- the patterns in variation were not consistent among study ligible evidence of a barred owl effect for Cle Elum, Tyee, areas. A cyclic, even-odd-year effect where fecundity was and Klamath study areas. The results of the meta-analysis high in even years and low in odd years was still important support much stronger negative effects of barred owl pres- for some study areas (Forsman et al. 2011), but has generally ence on spotted owl survival. The model with an additive become less evident since the last analysis (Anthony et al. barred owl effect ranked higher compared to the model with 2006). Overall, fecundity was declining in four areas (Cle an interaction between barred owl presence and study area, Elum, Klamath, South Cascades, Northwest California), supporting the importance of a consistent barred owl effect stable in two areas (Olympic, Tyee), and increasing in three across all study areas, rather than an effect that varies in areas (Rainier, Coast Ranges, H.J. Andrews) (table 2-4). magnitude among areas. The effects of several covariates on owl fecundity were The effects of climate, weather, and the amount of also reported by Forsman et al. (2011). The proportion of suitable owl habitat on survival were only investigated owl territories on each study area where barred owls were during the meta-analysis. There was some support for de- detected at least once during a breeding season had a nega- creasing time trends in survival and a negative relationship tive effect on fecundity for three study areas (Coast Ranges, between early nesting season precipitation and survival, Klamath, South Cascades), a positive effect on fecundity in but the amount of suitable habitat had no effect (Forsman one study area (H.J. Andrews), and no effect on the other et al. 2011). In addition, there was also some support for five areas. There was also evidence that low temperatures differences in survival among ecological regions, with the during the early nesting season had negative effects on fe- lowest survival rates reported for study areas in Washington cundity in three study areas (Rainier, Coast Ranges, South mixed-conifer regions and highest survival for the Coast Cascades); late nesting season temperatures had a negative Ranges. The meta-analysis suggested several factors effect on fecundity on one study area (Tyee); and high affected survival, but none of the covariates explored in precipitation during the early nesting season had negative this analysis explained a substantial portion of the variation effects on fecundity in three study areas (Cle Elum, Coast among years and study areas (between 0.0 and 5.7 percent Ranges, Northwest California). Support for a negative effect only). of barred owls and effects of climate and weather on fecundity was generally weak. In Oregon, increased fecundity Fecundity on four of five study areas (Coast Ranges, H.J. Andrews, The analysis across all 11 study areas by Forsman et al. Tyee, South Cascades) were associated with higher annual (2011) included 11,450 observations of the number of young estimates of the amount of suitable habitat associated with produced by territorial females, and 90 percent of those each study area; however, more suitable habitat resulted in observations were from adult females ( >3 years old). The decreased productivity on the Klamath study area (Forsman younger age classes were observed breeding much less fre- et al. 2011). There was little indication of any association quently (3.8 percent for 1-year-olds, 6.1 percent for 2-year- between the amount of suitable habitat and fecundity on olds), and age had a strong effect on productivity (Forsman the Washington study areas, and this association was not

12 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Table 2-3—Mean (x), age-specific fecundity (number of female young produced per female) with standard errors (SE) for northern spotted owls in the nine demographic study areas associated with land managed under the Northwest Forest Plan Age class Landowner 1 year old 2 years old ≥3 years old Study area class x SE x SE x SE Washington: Cle Eluma Mixed 0.115 0.083 0.517 0.109 0.553 0.052 Rainier Mixed 0.100 0.100 0.111 0.111 0.302 0.065 Olympica Federal 0.150 0.100 0.361 0.162 0.300 0.060 Oregon: Coast Rangesa Mixed 0.000 0.000 0.094 0.039 0.263 0.040 H.J. Andrewsa Federal 0.083 0.083 0.110 0.043 0.323 0.041 Tyeea Mixed 0.018 0.013 0.218 0.065 0.305 0.034 Klamatha Mixed 0.056 0.024 0.289 0.045 0.377 0.033 South Cascadesa Federal 0.060 0.038 0.210 0.064 0.347 0.052 California: Northwest Californiaa Mixed 0.088 0.054 0.152 0.038 0.324 0.027

Note: See table 2-1 for data years. a One of eight study areas monitored under the northern spotted owl effectiveness monitoring program for the Northwest Forest Plan. Source: adapted from Forsman et al. (2011).

Table 2-4—Trends in fecundity and survival, and mean rate of population change (x) with standard errors (SE) and 95-percent confidence limits (95% CI) for northern spotted owls from nine demographic study areas associated with land managed under the Northwest Forest Plan Estimated annual rate a of population change (λRJS) Landowner Population Study area class Fecundity Survival x SE 95% CI trendb Washington: Cle Elumc Mixed Declining Declining 0.937 0.014 0.910–0.964 Declining Rainier Mixed Increasing Declining 0.929 0.026 0.877–0.977 Declining Olympicc Federal Stable Declining 0.957 0.020 0.918 – 0.997 Declining Oregon: Coast Rangesc Mixed Increasing Declining since 1998 0.966 0.011 0.943–0.985 Declining H.J. Andrewsc Federal Increasing Declining since 1997 0.977 0.010 0.957–0.996 Declining Tyeec Mixed Stable Declining since 2000 0.996 0.020 0.957–1.035 Stationary Klamathc Mixed Declining Stable 0.990 0.014 0.962–1.017 Stationary South Cascadesc Federal Declining Declining since 2000 0.982 0.030 0.923–1.040 Stationary California: Northwest Californiac Federal Declining Declining 0.983 0.008 0.968–0.998 Declining a λRJS = reparameterized Jolly-Seber estimate of population change (Pradel 1996). b Population trends based on estimates of realized population change. c One of eight study areas monitored under the northern spotted owl effectiveness monitoring program for the Northwest Forest Plan. Source: adapted from Forsman et al. (2011).

13 GENERAL TECHNICAL REPORT PNW-GTR-850

investigated for California study areas because comparable et al. (2011) was 0.971 (SE = 0.007, 95-percent CI = 0.960 maps to develop the covariate were not available (Forsman to 0.983), which indicated that the average rate of popula- et al. 2011). tion decline was 2.9 percent per year during the study. The

weighted mean estimate of λ for the eight federal effective- Annual Rate of Population Change ness monitoring areas (excluding Rainier) was 0.972 (SE = Estimates of the annual rate of population change (λ) on 0.006, 95-percent CI = 0.958 to 0.985), which indicated an the nine study areas ranged from 0.929 to 0.996 (table 2-4). estimated decline of 2.8 percent per year. There was strong evidence that populations on the Cle Results from the meta-analysis on the annual rate of Elum, Rainier, Olympic, Coast Ranges, H.J. Andrews, and population change indicated that both survival and recruit- Northwest California study areas declined during the study ment differed by ecological region, with the highest survival (table 2-4, fig. 2-2), with particularly low estimates of λ for in the Oregon Coast Douglas-fir region and lowest survival Cle Elum and Rainier, which suggested population declines in Washington mixed-conifer region (Forsman et al. 2011). of 6.3 and 7.1 percent per year, respectively (table 2-4). Recruitment was highest in the Oregon/California mixed- Point estimates of λ for the Tyee, Klamath, and South Cas- conifer region and lower elsewhere (Forsman et al. 2011). cades study areas were all <1.0, but 95-percent confidence A negative association between barred owl detections and intervals (CIs) included 1.0 (table 2-4), suggesting popula- survival in the rate of population change analysis was also tions may be stationary. The weighted mean estimate of λ evident and consistent with results from the meta-analysis for all the study areas included in the analysis by Forsman of survival (see above). A weak association between sur-

Figure 2-2−Estimates of mean annual rate of population change (λ), with 95-percent confidence intervals for northern spotted owls in nine study areas associated with lands managed under the Northwest Forest Plan in Washington, Oregon, and California. Source: Forsman et al. (2011).

14 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

vival and the Pacific Decadal Oscillation was also evident, a decreasing trend in adult female survival. The annual rate with higher survival observed during warmer phases of this of population change was <1.0 for 10 of 11 areas examined, regional climate cycle. No other climate or weather covari- with an estimated average rate of population decline of ates were important. Estimates of recruitment were higher 4.5 percent per year (Burnham et al. 1996). By 2004, owl on study areas comprising primarily federal lands (Olympic, fecundity was relatively stable among the 14 study areas H.J. Andrews, South Cascades, Northwest California) examined, survival rates were declining on 5 of the 14 compared to mixed or private ownerships. Recruitment was areas, and populations were declining on 9 of 13 study areas also higher when the proportion of suitable owl habitat was for which there were adequate data to estimate λ (Anthony higher within study areas, but was lower in association with et al. 2006). However, the annual rate of decline was less, higher proportions of suitable habitat outside study area as mean λ�� for the 13 areas was 0.963, indicating popula- boundaries. tions were declining 3.7 percent annually during the study (Anthony et al. 2006). Realized Population Change Declines in fecundity, survival, and rate of population Estimates of realized population change reflected the trend change were observed across most study areas in this most in the proportion of the population remaining each year, recent analysis by Forsman et al. (2011). Over the last 15 based on annual changes in λ in relation to the popula- years, populations on all 11 areas included in the recent tion at the beginning of the study (Forsman et al. 2011). meta-analysis declined on average 2.9 percent per year Populations in Washington and northern Oregon (Olympic, (Forsman et al. 2011). This is a lower rate of decline than Rainier, Cle Elum, Coast Ranges) declined by 40 to 60 the 3.7 percent reported in the last meta-analysis (Anthony percent during this study (fig. 2-3), and there is some evi- et al. 2006), but the rates of decline are not directly com- dence that populations on the H.J. Andrews and Northwest parable between analyses. The current analysis represents California study areas were also declining (20 to 30 percent) a different time series than past efforts, and data collection although 95-percent CIs around estimates of realized on two of the study areas included in past analyses was population change overlapped 1.0 slightly (fig. 2-3). There discontinued (Wenatchee, Warm Springs Reservoir), so was less evidence that populations on South Cascades, these areas could not be included in the most recent analysis Tyee, and Klamath areas were in decline (5 to 15 percent), (Forsman et al. 2011). In addition to the Rainier study area, but many point estimates of realized population change for apparent survival rates of owls were declining on seven these areas were less than 1.0 even though 95-percent CIs (Cle Elum, Olympic, Coast Ranges, H.J. Andrews, Tyee, broadly overlapped 1.0 (fig. 2-3). South Cascades, Northwest California) of the eight study areas associated with the Plan (table 2-4) and fecundity Discussion was also declining in four of these populations (table 2-4) These demographic results are a summary of Forsman (Forsman et al. 2011). In Washington and northern Oregon, et al. (2011) and they represent the fifth meta-analysis of the number of declining populations and the rate of decline demographic data from northern spotted owls (Anderson raises concern about the long-term sustainability of the owl and Burnham 1992, Anthony et al. 2006, Burnham et al. throughout its range (Forsman et al. 2011). 1996, Franklin et al. 1999). The second meta-analysis of The reasons for declines in spotted owl populations demographic rates of northern spotted owls was conducted were not readily apparent in any of the previous meta- in 1993 and included 11 study areas (Burnham et al. 1996, analyses (Anthony et al. 2006, Burnham et al. 1996, Forsman et al. 1996). At that time, owl fecundity rates Franklin et al. 1999). The analysis done by Forsman et al. varied among years and with owl age, and exhibited no (2011) incorporated covariates to investigate the influence of increasing or decreasing trend over time (Burnham et al. barred owls, weather and climate, and habitat on fecundity, 1996). Survival rates were dependent on age, and there was survival, and rate of population change. As a result, we now

15 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 2-3−Estimates of realized population change, Δt, with 95-percent confidence intervals for northern spotted owls on nine study areas associated with lands managed under the Northwest Forest Plan in Washington, Oregon, and California, 1990–2006. 16 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

have some evidence that increasing numbers of barred owls Summary and loss of habitat contributed to demographic declines After 15 years of population monitoring, we continue to ob- reported in some study areas (Forsman et al. 2011). The serve significant annual declines in spotted owl populations presence of barred owls appeared to be the strongest and (2.9 percent all ownerships, 2.8 percent federal ownership) most consistent negative factor relating to spotted owl sur- (Forsman et al. 2011). Our ability to monitor the trend in vival, but the strength of the response was variable among owl populations is improving with newer technologies, the study areas. Forsman et al. (2011) concluded that although inclusion of explanatory covariates, and more years of data. their results do not represent cause-effect relationships, they We now have some evidence to support the suggestions of certainly suggest that barred owl invasion into the range of Anthony et al. (2006) that possible causes for declines in the spotted owl is at least partly to blame for the continued owl survival and populations may include high densities decline of the owl on federal lands. However, recovery of of barred owls and loss of habitat. However, a lot of uncer- habitat lost over the last century is a slow process and likely tainty remains, and we are just beginning to understand the continues to negatively impact owl populations. effects of these two factors on owl demography. We also From the perspective of evaluating the effectiveness must continue to stress the caution put forth in the Plan for of the Plan on the conservation and recovery of the owl, projecting current estimates of population decline into the the relationship between demographic rates and habitat future. are of particular importance. Because of the differences in At its implementation, the Plan’s assumption was that the vegetation data used to develop habitat models for the owl populations across the range would continue to decline 10-year report, as discussed in chapter 3 of this report, the for the first three to five decades, eventually stabilizing at development of the habitat covariate in California was not lower levels as losses of habitat lessen and habitat is restored possible, and its effect on demographic rates could only be in the network of large reserves scattered throughout its investigated for Washington and Oregon (see Forsman et al. range. Since the Plan’s inception, the rate of habitat loss has 2011 for details). From this analysis, there was evidence that certainly lessened, and here we report an overall habitat de- the percentage cover of suitable owl habitat had a positive cline of 3.4 percent on federal lands in the last 15 years (see influence on recruitment of owls in the meta-analysis of λ chapter 3 in this report), which is less than the anticipated (Forsman et al. 2011); however, this relationship was not rate of habitat loss of 5 percent per decade. We also report strong or prevalent for all demographic parameters or an overall 2.8 percent annual population decline on federal among all study areas. lands, with higher declines in the northern portions of the Based on the meta-analysis of λ, there was some range and stationary populations in the central portion of evidence that apparent survival was related positively to the the range as first noted by Anthony et al. (2006). These percentage cover of suitable habitat in the Cle Elum, Coast stationary populations were also not expected at the Plan’s Ranges, H.J. Andrews, and Tyee study areas in Washington implementation (Lint 2005). Although habitat is being and Oregon (Forsman et al. 2011). Also, a positive relation- maintained, the restoration of habitat under the Plan is still ship between recruitment and the percentage cover of suit- a few decades away. Forest succession is a slow process, able owl habitat within the study area in the meta-analysis but there are suggestions that it can be accelerated through of λ was also found (Forsman et al. 2011). Recruitment was well-designed silviculture (Garman et al. 2003, Muir et al. also highest on federally owned lands where the amount 2002). We were not yet able to accurately measure recruit- of suitable habitat was highest compared to private lands ment of nesting/roosting habitat with current technologies; (Davis and Lint 2005). One possible explanation for this however, we were able to detect recruitment of the younger result is that more suitable habitat within the study areas forests that serve as dispersal habitat (see chapter 3 in this provided areas where nonterritorial owls could survive until report). We speculate that declining spotted owl populations they were able to recruit into the territorial population.

17 GENERAL TECHNICAL REPORT PNW-GTR-850

will not begin to stabilize across the range at least until Anthony, R.G.; Forsman, E.D.; Franklin, A.B.; nesting/roosting habitat begins to increase significantly. Anderson, D.R.; Burnham, K.P.; White, G.C.; And although habitat is a key element in the conservation Schwarz, C.J.; Nichols, J.D.; Hines, J.E.; Olson, of spotted owls (Lint 2005), it may no longer be the primary G.S.; Ackers, S.H.; Andrews, L.W.; Biswell, B.L.; factor affecting population stability in either the short Carlson, P.C.; Diller, L.V.; Dugger, K.M.; Fehring, or long term. The rapidly increasing trend in barred owl K.E.; Fleming, T.L.; Gearhardt, R.P.; Gremel, populations has produced an unanticipated and confounding S.A.; Gutiérrez, R.J.; Happe, P.J.; Herter, D.R.; influence, as these species may compete for resources. Higley, J.M.; Horn, R.B.; Irwin, L.L.; Loschl, P.J.; The answer to the question, “Will the Plan reverse Reid, J.A.; Sovern, S.G. 2006. Status and trends in the declining population trend and maintain the historical demography of northern spotted owls, 1985–2003. geographic range of the northern spotted owl?” still eludes Wildlife Monographs No. 163. Washington, DC: The us. Five more years of monitoring has shed more light Wildlife Society. 48 p. on the subject, but a definitive answer will require more Burnham, K.P.; Anderson, D.R. 2002. Model selection long-term monitoring to better understand the temporal and multimodel inference: a practical information and spatial variability in owl demographics and the factors theoretical approach. 2nd ed. New York, NY: Springer- that affect owl vital rates. Until then, we believe that habitat Verlag. 488 p. maintenance and restoration, as currently envisioned under Burnham, K.P.; Anderson, D.R.; White, G.C. 1996. the Plan, remains essential to the owl’s recovery. However, Meta-analysis of vital rates of the northern spotted owl. additional conservation measures (i.e., barred owl control) Studies in Avian Biology. 17: 92–101. that were not envisioned under the Plan may ultimately be needed to recover the species in the face of the barred owl Courtney, S.P.; Blakesley, J.A.; Bigley, R.E.; Cody, expansion into the Pacific Northwest. M.L.; Dumbacher, J.P.; Fleischer, R.C.; Franklin, A.B.; Franklin, J.F.; Gutiérrez, R.J.; Marzluff, J.M.; References Sztukowski, L. 2004. Scientific evaluation of the status Anderson, D.R.; Burnham, K.P. 1992. Demographic of the northern spotted owl. Portland, OR: Sustainable analysis of northern spotted owl populations. In: Final Ecosystem Institute. 348 p. + appendixes. draft recovery plan for the northern spotted owl. Volume Davis, R.; Lint, J. 2005. Habitat status and trend. In: 2: 66–75. Unpublished report. On file with: U.S. Fish and Lint, J., tech. coord. Northwest Forest Plan—the first th Wildlife Service, 2600 SE 98 Ave., Suite 100, Portland, 10 years (1994–2003): status and trends of northern OR 97266. spotted owl populations and habitat. Gen. Tech. Rep. Anderson, D.R.; Burnham, K.P.; Franklin, A.B.; PNW-GTR-648. Portland, OR: U.S. Department of Gutiérrez, R.J.; Forsman, E.D.; Anthony, R.G.; Agriculture, Forest Service, Pacific Northwest Research White, G.C.; Shenk, T.M. 1999. A protocol for conflict Station: 21–82. resolution in analyzing empirical data related to natural Fleming, T.L.; Halverson, J.L.; Buchanan, J.B. 1996. resource controversies. Wildlife Society Bulletin 27: Use of DNA analysis to identify sex of northern spotted 1050–1058. owls (Strix occidentalis caurina). Journal of Raptor Research. 30: 18–122.

18 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Forsman, E.D.; Anthony, R.G.; Dugger, K.M.; Glenn, Franklin, A.B.; Gutíerrez, R.J.; Nichols, J.D.; Seamans, E.M.; Franklin, A.B.; White, G.C.; Schwarz, C.J.; M.E.; White, G.C.; Zimmerman, G.S.; Hines, Burnham, K.P.; Anderson, D.R.; Nichols, J.D.; Hines, J.E.; Munton, T.E.; LaHaye, W.S.; Blakesley, J.E.; Lint, J.B.; Davis, R.J.; Ackers, S.H.; Andrews, J.A.; Steger, G.N.; Noon, B.R.; Shaw, D.W.H.; L.S.; Biswell, B.L.; Carlson, P.C.; Diller, L.V.; Keane, J.J.; McDonald, T.L.; Britting, S. 2004. Gremel, S.A.; Herter, D.R.; Higley, J.M.; Horn, R.B.; Population dynamics of the California spotted owl Reid, J.A.; Rockweit, J.; Schaberl, J.; Snetsinger, T.J.; (Strix occidentalis occidentalis): a meta-analysis. Sovern, S.G. 2011. Population demography of northern Ornithological Monographs. 54 p. spotted owls. Studies in Avian Biology 40. Berkeley, CA: Franklin, A.B.; Gutiérrez, R.J.; Noon, B.R.; Ward, University of California Press. 106 p. J.P. 1996b. Demographic characteristics and trends Forsman, E.D.; Anthony, R.G.; Reid, J.A.; Loschl, P.J.; of northern spotted owl populations in northwestern Sovern, S.G.; Taylor, M.; Biswell, B.L.; Ellingson, A.; California. Studies in Avian Biology. 17: 83–91. Meslow, E.C.; Miller, G.S.; Swindle, K.A.; Thrailkill, Garman, S.L.; Cissel, J.H.; Mayo, J.H. 2003. J.A.; Wagner, F.F.; Seaman, D.E. 2002. Natal and Accelerating development of late-successional conditions breeding dispersal of northern spotted owls. Wildlife in young managed Douglas-fir stands: a simulation Monograph No. 149. Washington, DC: The Wildlife study. Gen. Tech. Rep. PNW-GTR-557. Portland, OR: Society. 35 p. U.S. Department of Agriculture, Forest Service, Pacific Forsman, E.D.; DeStefano, S.; Raphael, M.G.; Northwest Research Station. 119 p. Gutíerrez, R.J., eds. 1996. Demography of the northern Lebreton, J-D.; Burnham, K.P.; Clobert, J.; Anderson, spotted owl. Studies in Avian Biology No. 17. Camarillo, D.R. 1992. Modeling survival and testing biological CA: Cooper Ornithological Society. 122 p. hypotheses using marked animals: a unified approach Franklin, A.B.; Anderson, D.R.; Forsman, E.D.; with case studies. Ecological Monographs. 62: 67–118. Burnham, K.P.; Wagner, F.F. 1996a. Methods for Lint, J.; Noon, B.; Anthony, R.; Forsman, E.; Raphael, collecting and analyzing demographic data on northern M.; Collopy, M.; Starkey, E. 1999. Northern spotted spotted owl. Studies in Avian Biology. 17: 12–20. owl effectiveness monitoring plan for the Northwest Franklin, A.B.; Burnham, K.P.; White, G.C.; Anthony, Forest Plan. Gen. Tech. Rep. PNW-GTR-440. Portland, R.G.; Forsman, E.D.; Schwarz, C.; Nichols, J.D.; OR: U.S. Department of Agriculture, Forest Service, Hines, J. 1999. Range-wide status and trends in north- Pacific Northwest Research Station. 43 p. ern spotted owl populations. Colorado Cooperative Fish Lint, J.B., tech. coord. 2005. Northwest Forest Plan—the and Wildlife Research Unit, Colorado State University, first 10 years (1994–2003): status and trends of northern Fort Collins, Colorado, USA. 72 p. http://warnercnr. spotted owl populations and habitat. Gen. Tech. Rep. colostate.edu/~alanf/reprints/abf_ea99nsometa.pdf. PNW-GTR-648. Portland, OR: U.S. Department of (18 December 2010). Agriculture, Forest Service, Pacific Northwest Research Station. 176 p.

19 GENERAL TECHNICAL REPORT PNW-GTR-850

Muir, P.S.; Mattingly, R.L.; Tappeiner, J.C., II; Bailey, Pradel, R. 1996. Utilization of capture-mark-recapture for J.D.; Elliott, W.E.; Hagar, J.C.; Miller, J.C.; Peterson, the study of recruitment and population growth rate. E.B.; Starkey, E.E. 2002. Managing for biodiversity Biometrics. 52: 703–709. in young Douglas-fir forests of western Oregon. U.S. SAS Institute, Inc. 2008. SAS/STAT® 9.2 users guide. Geological Survey, Biological Resources Division, SAS Institute, Inc., Cary, NC. Biological Science Report USGS/BRD/BSR–2002-0006. 76 p. http://fresc.usgs.gov/products/papers/mang_bio.pdf. White, G.C.; Burnham, K.P. 1999. Program MARK: (18 December 2010). survival estimation from populations of marked animals. Bird Study. 46: 120–138. Peter Carlson

Participants in the 2009 META-Analysis.

20 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Chapter 3: Habitat Status and Trend Raymond J. Davis and Katie M. Dugger Introduction and Classification and Assessment with Landsat of Visible The first rangewide northern spotted owl Strix( occidentalis Ecological Groupings (CALVEG) data (California) (Davis caurina) habitat map was developed for the Forest Eco- and Lint 2005, Moeur et al. 2005). The choice of vegetation system Management Assessment Team (FEMAT) in 1993. variables provided by these two sources was limited and It was constructed through a combination of digital maps included only tree size class and cover attributes, which derived from satellite imagery and maps derived from aerial were not mapped consistently between the two products. photo interpretation. The team used the best available data Other habitat mapping “core elements” discussed by Lint and geographical information system (GIS) technologies et al. (1999), such as stand age and tree species data, were at that time to represent owl habitat conditions at the start not available, resulting in omission of important habitat of the Northwest Forest Plan (the Plan), which we call the relationship variables in the models used to create the first “baseline.” However, the authors acknowledged that the baseline map. To compensate for lack of tree species data, map was an estimate and had not been assessed for ac- Davis and Lint (2005) used elevation as a variable in their curacy (FEMAT 1993). Six years later, the northern spotted habitat modeling and also built a “habitat-capable” GIS owl effectiveness monitoring plan concluded that this map layer, largely based on a rangewide elevation isopleth that lacked the spatial resolution and accuracy needed for a would “mask out” subalpine forests, in which spotted owls baseline spotted owl habitat map for monitoring purposes avoid nesting. There was no way to “mask” pine-dominated (Lint et al. 1999). They proposed the development of a new forests or to include evergreen hardwoods, which are rangewide baseline habitat map to “provide the landscape- important components of owl habitat in the southern phys- scale view of habitat conditions at different resolutions.” iographic provinces. As a result, where tree size and cover Having a good baseline habitat map is essential to the conditions were otherwise similar to those used by nesting effectiveness monitoring program because it provides a and roosting territorial owls, the models classified them as snapshot in time of what conditions were like when the suitable, even when they probably were not because of tree Plan was implemented. Without an understanding of base- species composition. line conditions, we would not be able to answer the primary Another problem was the coarse spatial resolution and question of whether owl habitat and dispersal habitat are lack of continuous attribution in the CALVEG data (Davis being maintained and restored under the Plan. The first and Lint 2005). This resulted in poorer estimates of habitat rangewide baseline habitat monitoring map was developed in the California physiographic provinces and habitat maps by Davis and Lint (2005) for the 10-year monitoring report that were not directly comparable to the Oregon and Wash- (Lint 2005). The data sources and methods used to develop ington maps. The lack of a consistent rangewide habitat map that map are fully described in Davis and Lint (2005) and resulted in our inability to fully model associations between are not repeated in this report. Limitations in the first base- spotted owl demography rates and habitat during the 2009 line map were noted by Davis and Lint (2005) and Raphael population meta-analysis (Forsman et al. 2011). (2006) and are reviewed in the following discussion. Additional limitations of the 10-year report’s baseline The Northwest Forest Plan effectiveness monitoring owl habitat map (Davis and Lint 2005) included the use of program was in its early stages of development at the time the median algorithm in the BioMapper habitat modeling of the 10-year reporting analysis. A consistent rangewide software (Hirzel et al. 2002), which was the only algorithm vegetation data set as described in Lint et al. (1999) did available at the time (Davis and Lint 2005). This algorithm not exist. Instead, two distinctly different vegetation data assumed species distribution along the environmental sources covered the owl’s range: Interagency Vegetation factors was normal (see fig. 3-8 on page 36 in the 10-year Mapping Project (IVMP) data (Oregon and Washington) report); however, in reality, this is not always the case, and nonnormal relationships resulted in the overestimations of

21 GENERAL TECHNICAL REPORT PNW-GTR-850

habitat suitability. In general, profile models like BioMap- California. This marks the first application of using multiple per are known to sometimes overpredict habitat suitability satellite imagery dates to create “bookend” vegetation maps (Engler et al. 2004). To compensate for this, Davis and Lint for habitat monitoring purposes (Ohmann et al. 2010). (2005) provided a habitat map with a continuous scale from In addition to improved vegetation map products, the 0 to 100, where a value close to zero signified that an indi- science of habitat modeling has evolved since the 10-year vidual map unit (pixel) had little in common with the condi- report. Species distribution and habitat suitability modeling tions found where territorial owls are present, and those has been the subject of much current research and discus- with values close to 100 had much in common with sites sion in ecology (Elith et al. 2006, Guisan and Zimmermann having territorial owl presence. During this initial effort, a 2000, Hirzel and Le Lay 2008), so we spent a substantial threshold value that designated a cutoff between “suitable” amount of time reviewing modeling options and testing and “not suitable” habitat was not chosen. Instead, Davis several types of software used for habitat modeling before and Lint (2005) reported on status and trend of the spectrum deciding on the approach presented here. of habitat suitability (HS) divided into equal-interval bins, One thing we have observed through these efforts is and areas with HS >40, which “had characteristics similar that regardless of the methods used, the map products are to areas where territorial owls have been found.” visually similar at the rangewide scale (fig. 3-1). Therefore, Based on our latest work (presented here), we now it is important to test the map’s accuracy with actual spotted conclude that the baseline habitat map developed for the owl nesting and roosting location data. This is one area 10-year report did overestimate owl habitat suitability in where the population monitoring and habitat monitoring portions of the range. Overestimations occurred within efforts connect, as we used different subsets of the demo- pine-dominated forests of the eastern Cascades for reasons graphic data to first train and then test the accuracy of our discussed above, and, as noted by Raphael (2006), habitat habitat model mapped predictions. suitability scores greater than 40 were achieved in stands The use of the new rangewide vegetation data set and as young as 30 years in the Coast Range of Oregon and the latest habitat modeling software has resulted in an 50 years in Oregon western Cascades, providing further improved baseline habitat map that has tested well with evidence of profile model overpredictions. Based on visual actual owl pair location data (including independent data comparisons of the former baseline maps and the new one, sets). These improvements included better discrimination we also believe that the use of the coarser scale CALVEG of habitat in the eastern Cascades, where pine-dominated data in the 10-year habitat modeling resulted in consider- forests mostly occur, and the use of the “habitat-capable” ably more habitat suitability >40 estimated for California. layer from Davis and Lint (2005) was no longer required Since the 10-year report, much progress has been made for habitat modeling with the inclusion of a subalpine forest in developing a consistent rangewide vegetation data layer, type variable. We use this new baseline map (1994/96) and with a larger suite of vegetation attributes to be used as the other bookend map (2006/07) for conducting our habitat “core elements” for habitat mapping, including tree spe- status and trend analysis. cies information (Ohmann and Gregory 2002). These new The development of bookend maps was an innovative rangewide vegetation data are produced by the Landscape advancement in our monitoring methods, but aspects of Ecology, Modeling, Mapping, and Assessment group it remain to be tested. Given its novelty, we restricted our (LEMMA) based at the Pacific Northwest Research Station use of the 2006/07 bookend to only inform us on habitat in Corvallis, Oregon (link to Web page: http://www.fsl.orst. changes within areas that were identified as having experi- edu/lemma/). Detailed attributes of forest composition and enced a disturbance by the LandTrendr data. It is important structure were mapped for all forests in the Plan area for to make sure that the bookend maps used for later analyses two “bookend” dates. The bookend dates were 1996 and are generated with the same data sets and methods, and 2006 in Washington and Oregon, and 1994 and 2007 in tested so that the detection of change from one to the other

22 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3-1—Baseline maps are representative of habitat at the implementation of the Northwest Forest Plan. They evolve with new mapping technology. Even so, at the range scale, the general spatial patterns of habitat between them are similar. FEMAT = Forest Ecosystem Management Assessment Team. is a fair comparison of “real” change and not one caused is to determine if assumptions made during the develop- by analytical or data differences. In future monitoring ment of the Plan are holding true. Testing the assumption cycles, we anticipate more advancements in both vegetation that habitat will not decline faster than predicted in the final data mapping and habitat modeling science; therefore, we environmental impact statement (USDA and USDI 1994) anticipate that future modifications will be made to the is of particular interest. The initial list of assumptions is as baseline map, including the use of 1994 satellite imagery for follows (Lint et al. 1999): the entire range. This is appropriate as the status and trend 1. Habitat conditions within late-successional reserves analysis is based on the use of the best available vegetation (LSRs) will improve over time at a rate controlled by and change-detection data and technologies. successional processes in stands that currently are not Habitat Monitoring Under the Plan habitat. However, this is not expected to produce any significant changes in habitat conditions for several Under the monitoring plan, habitat status and trends are to decades. be estimated approximately every 5 years after the baseline map was developed because it was believed that changes 2. Habitat conditions outside of reserved allocations in forest vegetation conditions would not be discernable will generally decline because of timber harvest and from the remote sensed vegetation data on more frequent other habitat-altering disturbances, but the vegetation intervals (Lint et al. 1999). The intent of habitat monitoring structure across the landscape will continue to facilitate owl movements.

23 GENERAL TECHNICAL REPORT PNW-GTR-850

3. Catastrophic events are expected to halt or reverse report status and trends within every land use allocation. the trend of habitat improvement in some reserves; Instead, we will report by broad federal land use allocations however, the repetitive design of reserves should representing “reserved” and “nonreserved” landscapes (fig. provide resiliency, and not result in isolation of 3-3), which we feel is a more consistent and appropriate population segments. scale for monitoring. Because the “large block” reserves Central to these questions is the federal network of (see fig. 3-13, page 44 in the 10-year report) make up about reserved land use allocations designed to support groups 90 percent of the reserved landscape, we now consider our of reproducing owl pairs across the species’ range. These reporting of status and trend in the reserved landscape as reserves include LSRS, adaptive management reserves, one entity, whereas in the 10-year report we separated them. congressionally reserved lands, managed late-successional Although the effectiveness monitoring is focused to address areas, and larger blocks of administratively withdrawn questions about the Plan, its developers realized that the lands. It is also important to monitor the lands between status and trends of the subjects being monitored are often these reserves because they provide for recruitment of new influenced by conditions on the surrounding nonfederal owls into the territorial populations (see chapter 2, this re- lands. Therefore, we will report on habitat conditions on port) and are important for dispersal and movement of owls nonfederal lands at the state and range scales because these between larger reserves. These dispersal habitats occur in a were included in the 10-year monitoring synthesis report by combination of matrix, adaptive management areas, riparian Raphael (2006). reserves, small tracts of administratively withdrawn lands, As stated in the 10-year report, our objective was to and other small reserved areas such as 100-ac owl core produce maps of forest stands (regardless of patch size and areas. To understand whether the Plan is contributing to the spatial configuration) that showed the level of similarity to conservation and restoration of owl habitat, the condition stand conditions known to be used for nesting and roosting and trends of owl habitat must be regularly assessed. The by spotted owls. Forest stands with conditions most similar specific questions that were addressed in the 10-year report to what is used by nesting and roosting owl pairs are what and that will be addressed here as well include: we will refer to as “nesting/roosting habitat” throughout this document. We will also report on forest stand conditions 1. What proportion of the total landscape on federal lands that are known to be used by dispersing owls, which we are owl habitat and dispersal habitat? refer to as “dispersal habitat.” 2. What are the trends in amount and changes in distribution of owl habitat, particularly in large, Methods and Data Sources reserved blocks? Land Use Allocation Data 3. What are the trends in amount and distribution of An updated map of the Plan’s land use allocations (LUA) dispersal habitat outside of the large, reserved blocks? was produced in 2002 for the 10-year monitoring reports (Huff et al. 2005, Lint 2005, Moeur et al. 2005). It updated 4. What are the primary factors leading to loss and the original 1994 version, which was mapped with older fragmentation of both owl habitat and dispersal habitat? GIS technology and had a 40-ac resolution. This first Following the approach of Davis and Lint (2005), the update corrected some mapping inconsistencies, but more condition of owl habitat will be reported at three broad geo- importantly, incorporated allocation changes that occurred graphic scales: (1) the physiographic province, (2) the state, between 1994 and 2002. Although this map was considered and (3) the geographic range of the owl. However, because an improvement from the earlier version, some limitations of changes that have occurred in federal land use alloca- still remained (Davis and Lint 2005, Huff et al. 2005). The tions since the 10-year report (fig. 3-2), we will no longer major limitations were the inability to map riparian reserves

24 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3.2—Changes made to the land use allocations since the 10-year report (Lint et al. 2005).

25 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 3-3—Federally administered lands within the range of the northern spotted owl.

26 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

(which can cover significant amounts of land where stream improving it for each monitoring effort. For monitoring densities are high) and inconsistencies in how administra- purposes, we archive the previous versions and report tively withdrawn areas were delineated. Errors that re- vegetation and habitat changes for all monitoring modules mained after the 2002 update included the misidentification within the reference frame of the most up-to-date alloca- of a state-owned park in the redwood region of California as tion map. Major LUA changes that are important for us federally owned National Park Service land and inaccurate to note include changes that cover thousands of acres and or missing boundaries of national wildlife refuges, mainly involve gains or losses of reserved allocations. We will in Washington and Oregon. Other minor mapping issues discuss these changes in relationship to the standard and included edge matching inconsistencies that caused “sliver guidelines within the record of decision (USDA and USDI gaps” and inconsistent attribution of large water bodies. 1994). Given the most recent information, the latest changes A second update of LUAs performed in 2009 (fig. 3-2) in reserved allocations (fig. 3-2) have resulted in a slightly produced a new version that is used for this 15-year report. increased area and improved distribution and connectivity The new version incorporates major LUA changes that of the reserved allocation system. occurred between 2002 and 2009, and it also corrects the errors identified above. Minor issues with inconsistent map- Vegetation Data ping of administratively withdrawn areas still remain, and The vegetation data used for habitat modeling and mapping a small amount (<1 percent) of federally administered lands were developed through a method for predictive vegetation are awaiting official LUA designations and are identified as mapping using direct gradient analysis (Gauch 1982, ter “not designated” in the 2009 map. Riparian reserves still Braak 1986) and nearest-neighbor imputation (Moeur and remain unmapped because, as Moeur et al. (2005) noted, Stage 1995) to assign detailed forest vegetation plot infor- “…at the Plan scale, they cannot be reliably distinguished mation to every pixel in a GIS raster map. The combining from matrix because of a lack of consistency in defining of these methods to develop vegetation maps was termed intermittent stream corridors and varying definitions for “gradient nearest neighbor” (GNN) and is thoroughly riparian buffers.” described in Ohmann and Gregory (2002). The GNN maps The Plan allowed for land exchanges involving LSRs developed in the Pacific Northwest have previously been if they provide benefits equal to or better than current applied to broad-scale vegetation mapping efforts across a conditions, such as to improve area, distribution, and wide range of forest ecosystems (Ohmann et al. 2007, Pierce connectivity of the LSR system (USDA and USDI 1994). et al. 2009). Forest attributes from regional inventory plots It also acknowledges that future changes would occur for are assigned to map pixels where data are missing, on the the administratively withdrawn allocation. At the end of basis of a modeled relationship between the detailed forest the 15-year monitoring period, we note a net increase of attributes from plots and a combination of spatial predictor about 25,000 ac in reserved allocations, and a net decrease variables derived from Landsat satellite imagery, climate of about 17,000 ac in nonreserved allocations. Most of variables, topographic variables, and soil parent materials. the changes that occurred were designations of otherwise The assumption behind GNN methods is that two locations reserved allocations into 237,000 ac of congressionally des- with similar combined spatial “signatures” should also have ignated reserves, and most of this (83 percent) occurred in similar forest structure and composition. Plot data are from northern California. Because some of the changes included regional forest inventory plots: Forest Inventory and Analy- land exchanges or acquisitions, the increase in reserved sis (FIA) periodic inventories on nonfederal lands, FIA allocations and decrease in nonreserved allocations are not annual inventory on all ownerships, and Current Vegetation equal. Survey inventories. The GNN data used for habitat model- Land use allocations will continue to change, and ing and mapping covers the entire breadth of the owl’s range we will continue to update this map with the intent of from Washington to northern California for two points in

27 GENERAL TECHNICAL REPORT PNW-GTR-850

time. We call these two data sets “bookends” because the for these species groups, or forest-type, variables ranged changes in habitat that we analyzed and report on occurred from 0.30 to 0.46, with an average kappa of 0.40 (±1 SD = between them. The satellite imagery from which GNN was 0.07). Oak woodland was the most accurate species group, created covers the period from 1994 to 2007 in California followed by subalpine, evergreen hardwoods, and pine. and 1996 to 2006 in Oregon and Washington. The on-the- ground plot data used to create the vegetation maps covers Change-Detection Data the period 1991 to 2000 for bookend 1, and 2001 to 2007 for A new approach to monitoring landscape vegetation change bookend 2. The GNN products are 30-m (98.4-ft) grids that was implemented to map forest disturbances in the owl’s were specifically developed for mid- to large-scale spatial range. Landsat-based detection of trends in disturbance analysis (Ohmann and Gregory 2002). and recovery (LandTrendr) produces yearly maps of The primary challenge was to develop GNN model- forest disturbance using a new analysis of annual Landsat based maps for the two bookend dates that minimized Thematic Mapper satellite imagery (Kennedy et al. 2010). spectral differences owing to different image dates that In general, LandTrendr detects spectral trajectories from might produce false vegetation changes. To achieve this, the Landsat time-series stacks and correlates them to land GNN models used Landsat imagery that was geometrically surface changes. The time series of Landsat imagery that rectified and radiometrically normalized through time was assembled for the Plan area was processed using basic using the LandTrendr algorithms (Kennedy et al. 2007). A atmospheric correction, cloud screening, and radiometric full description of the GNN bookends methodology can be normalization to separate imagery noise (i.e., cloud cover, found in Moeur et al. (2011). smoke, snow, or shadows) from actual vegetation change. The accuracy assessment for GNN continuous vari- Predictions of vegetation cover change were then evaluated ables was based on the correlation of observed plot values using a statistical model of vegetation cover developed from against predicted (modeled) values. Ohmann et al. (2010) photointerpreted plots (Cohen et al. 2010). The results of used a modified leave-one-out cross-validation approach this evaluation found that LandTrendr detected vegetation that yields results similar to those of a true cross-validation disturbances as well as or better than two-date change- approach, but probably slightly underestimates the true detection methods, and that it detects with reasonable accuracy. The accuracy assessments are based on pooled robustness a range of other dynamics such as insect-related plots for each modeling region. Canopy characteristics are disturbance and growth (Kennedy et al. 2010). Errors in usually the most easily determined via space-borne remote LandTrendr predictions were generally confined to very sensing instruments, and the most accurate GNN variable subtle change phenomena (Kennedy et al. 2010). In sum- was conifer canopy cover, with an average plot correla- mary, LandTrendr improved the temporal frequency of tion of 0.74 (±1 standard deviation [SD] = 0.07). Inferring disturbance maps used for monitoring, better separates vegetation characteristics underneath the canopy is more subtle changes from background noise, and detects a wider difficult, and the correlation coefficients for the structural range of vegetation change phenomena than was possible and age vegetation variables we chose to use ranged from with previous technologies (Kennedy et al. 2010, Moeur et 0.38 to 0.82, with an average plot correlation of 0.63 (±1 SD al. 2011). = 0.12). The accuracy assessment for the species composi- We used the LandTrendr data to verify habitat losses tion variables is based on Cohen’s kappa coefficient, which between our bookend maps and to attribute the most likely is a measure of agreement between predicted and actual cause of habitat loss (fig. 3-4). The data covered the entire conditions (in this case dominant tree species), taking into analysis area and period (1994–2007) and provided infor- consideration agreement occurring by chance (Cohen 1960). mation by 30- by 30-m pixels on initial year of disturbance. We combined several species to produce “forest type” basal LandTrendr classified the cause of disturbance (vegetation area variables as shown in appendix A. The average kappas cover loss) into three types: (1) timber harvest, (2) insect

28 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3-4—LandTrendr change-detection data (Kennedy et al. 2010).

29 GENERAL TECHNICAL REPORT PNW-GTR-850

and disease (can also include pathogens and other nonabrupt contain all the environmental variables that were used for processes), and (3) wildfire. Fire locations were identified habitat modeling. Our solution was to supplement the owl based on fire perimeter GIS data from Monitoring Trends location data from demography study areas with the owl in Burn Severity (MTBS1) data, Geospatial Multi-Agency presence location from the broader geographic areas sur- Coordination (GeoMAC2) data, and other sources (i.e., indi- rounding them to reduce sampling bias issues and produce vidual forest data). The remaining short-term disturbances a training data set that was better distributed within the were assigned “harvest” as the probable cause of distur- modeling region. To do this we used the data set used for bance, although wind may account for a small percentage. the 10-year report (Davis and Lint 2005). The first step in this process was to conduct a nearest Spotted Owl Presence Data neighbor distance analysis on owl pair site centers from The owl survey data collected under the effectiveness study areas within each modeling region (app. B). We used monitoring program are important not only for population the average nearest neighbor distances calculated from the monitoring (chapter 2, this report), but also for monitoring 50-percentile harmonic cores (to remove outlier sites) from suitable habitat. The owl pair location data (presence only each of the study areas as a minimum distance parameter spatial data) from demographic study areas are collected for randomly selecting a number (equal in size to the demo- annually and are spatially very accurate. This made such graphic study area data) of northern spotted owl pair sites data ideal for habitat suitability modeling; thus, we used from the 10-year report training data set (Davis and Lint them as the foundation for training our habitat models. 2005) that was outside of the study area boundaries. Both However, the results of our preliminary model testing the demographic study area sites and the random selection indicated that using only demographic study area data was of owl pair locations outside of them were combined to form problematic for modeling habitat in some modeling regions. the habitat suitability model training data set. This provided Confining our model training data to the demography areas a well-distributed and nonclumped training data set for each produced a “geographically clumped” distribution of model modeling region. training points within the boundaries of our larger modeling We also attempted to match the date of our training regions. This clumping violated the basic assumption of data to the date of the satellite imagery used to create habitat modeling methods that require independence and the vegetation data set that provided habitat variables for sampling without bias for presence data (training data) from modeling. And finally, because we suspected interspecific the modeling region (Gillison and Brewer 1985, Phillips et competition between spotted owls and barred owls (Strix al. 2009, Williams et al. 2002). We therefore matched our varia) to potentially confound the spotted owl/habitat use modeling regions to the boundaries of the demographic relationship, we used activity centers from the study areas study areas, trained the habitat model to those areas, and based on surveys done between 1994 and 1996 because then extrapolated the model results to the larger geographic barred owl densities were lower than in 2006 and 2007. regions. This produced mixed results, with some models Our training data outside of the study area cover a broader testing well, while others could not be projected (extrapo- period that roughly frames that period as discussed in Davis lated) successfully when the larger geographic area did not and Lint (2005).

1 Data accessible thru the Forest Service’s Remote Sensing Ap- Habitats, the Niche Concept, and Habitat plications Center (RSAC) Web site http://www.fs.fed.us/eng/rsac/. Modeling 2 Geospatial Multi-Agency Coordination Group or GeoMAC, is an Internet-based mapping application originally designed for Understanding where animals live and the myriad factors fire managers to access online maps of current fire locations and perimeters in the conterminous 48 states and Alaska. Data are associated with how and why they make the choices as available at http://www.geomac.gov/. to where to live has been the subject of extensive research (Stauffer 2002). As stated by Morrison et al. (1992), “an

30 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

animal’s habitat is, in the most general sense, the place Hirzel (the developer of BioMapper) with known species where it lives.” This seems simple enough: an animal occurrence and distributions. The details of these tests are can only live in an area that meets its basic needs for re- not provided in this report; however, our conclusions were sources (food, water, nest sites, etc.), includes competitors similar to those of Braunisch and Suchant (2010) who found and predators with which it can coexist, and in climatic that BioMapper and MaxEnt produced models with similar extremes it can withstand (Morrison et al. 1992). This is accuracy, but that MaxEnt performed better when trained maybe best articulated within the niche concept, which has with systematically sampled data that were well-distributed a long evolution in the science literature (see Morrison et within the modeling region. However, BioMapper outper- al. 1992 for review) and has become a useful construct for formed MaxEnt when the model results were extrapolated conceptualizing and quantifying wildlife-habitat relation- to areas outside of the model training data area (Braunisch ships. The multivariate, or n-dimensional, niche as defined and Suchant 2010). In summary, our tests found that as by Hutchinson (1957) lends itself well to current attempts long as species presence [training] data were fairly well to model wildlife-habitat interactions, as it allows us to distributed within a modeled region, MaxEnt outperformed conceptualize all the complexities associated with how and the other modeling methods, and we selected it as the why animals choose where they live. A species potential or habitat modeling tool for this reporting cycle. Several other “fundamental” niche includes a subset of all the environ- comparisons between MaxEnt and a number of other habitat mental conditions required for a species long-term survival; modeling approaches are available in the scientific litera- however, this “fundamental niche” can be further restricted ture, and in most cases, distribution models generated by by predators and competitors resulting in a “realized niche” MaxEnt performed as well or better than the other methods (Hutchinson 1957). This realized niche reflects a subset of (Baldwin 2009). the conditions found in the fundamental niche and is the set Other notable factors associated with our selection of of environmental conditions that characterize the space a MaxEnt included its user-friendly interface, its ability to species actually occupies (Hutchinson 1957) and is reflected run replicated models for testing purposes and to provide in the observed distribution of a species. information on the importance of the environmental vari- Many types of species distribution models are available ables used for modeling, and most importantly, its ability to for estimating a species’ realized niche (and producing a “project” or “transfer” model results. Model transferability geographic distribution map of it) using species presence is the term given for applying the results of a model that data that are correlated to environmental data of relevance is calibrated for specific location or period, to a different to the species occurrence. For the 10-year report (Davis and geographic location or period (Turner et al. 1989). The Lint 2005), we used modeling software called BioMapper concept is based on the idea that calibrated model param- (Hirzel et al. 2002). However, species distribution modeling eters from one area or time may provide useful information is a rapidly evolving field of study, so before conducting the in estimating conditions in a different time or place. In our spotted owl habitat modeling, we conferred with some of situation, we attempted to transfer our models, which were the species distribution modeling software developers (A. trained in 1994/96 to the same geographic location, but in a Hirzel and S. Phillips) and evaluated various habitat model- different period—2006/07. Model transferability is a fairly ing methods (i.e., BioMapper: Hirzel et al. 2002; MaxEnt: new concept, and one that is rarely assessed (Randin et Phillips et al. 2006, Phillips and Dudík 2008; Mahalanobis al. 2006). Issues with MaxEnt projections documented by distance method: Jenness 2003; resource selection func- Braunisch and Suchant (2010), our model testing, and the tions: Manley et al. 1993). We also ran comparison tests current literature advise for caution in its use and interpreta- between BioMapper (all algorithms) and MaxEnt using tion (Jiménez-Valverde et al. 2009, Peterson et al. 2007, “virtual species” data sets provided by Dr. Alexandre Phillips 2008).

31 GENERAL TECHNICAL REPORT PNW-GTR-850

Habitat Modeling Process feature” that performs a function similar to Akaike’s MaxEnt uses a machine learning process and a suite of information criterion (Akaike 1974) by penalizing the com- potential response functions to estimate the most uniform plexity of the model. The regularization multiplier affects distribution (maximum entropy) of the “average” environ- the fit of the model training data to the modeling variable mental conditions at known species locations compared empirical means. A smaller value results in a tighter fit but to what is available across the modeled area (background) potentially leads to overfitting the model to the data. The (Phillips et al. 2006). The modeling process does not default setting of 1.0 is believed to be an appropriate setting require an a priori specification of a set of models, but for most modeling efforts (Phillips and Dudík 2008). A instead fits training data (presence locations of owl pairs) higher regularization multiplier setting reduces the number to environmental covariates using various combinations of model parameters, allowing for a more spread out fit of response functions (features) such as linear, quadratic, around the mean, and simplifies the model. product, hinge, and threshold structures. However, the use Observing the statistical performance on test (versus of all feature types may lead to model overfitting depending training) data is the best approach to final model calibration on the sample size of the training data (Phillips et al. 2006); (Phillips 2010). We therefore evaluated our model’s perfor- therefore, the “auto feature” (default) restricts the model to mance beginning with the model test gain, which indicates simpler features, such as linear, quadratic, and hinge, for how different the testing data are from the background data. smaller sample sizes (Elith et al. 2011). In our preliminary It is similar to “deviance” as used in generalized linear model tests, overfitting seemed to occur from the use of the modeling (Phillips et al. 2006) and higher gains indicate threshold feature, which requires a minimum of 80 training larger differences between occurrence location environ- samples and produced sharp jumps (both up and down) in mental conditions and average background environmental the variable response curves. Modeling with just the hinge conditions. The exponent of gain produces the mean pro- feature produces models with simpler or smoother functions bability value of predicted species occurrence compared to and is generally a useful simplification that can reduce a random location selected from the surrounding modeled overfitting (Phillips 2010). Our final selection incorporated landscape. Or in other words, an average testing gain of a combination of linear, product, and hinge features because 0.80 indicates that the model predicted owl occurrence most of our hypothesized variable responses fit those 2.2 times what would be expected by chance. In addition, choices. We considered using the quadratic feature; how- observing the differences between model testing gain and ever, during our model testing, MaxEnt applied this feature regularized training gain can be used to control model over- to variables in which the response function did not make fitting, as a large difference between the two is an indication ecological sense (i.e., tree diameter). This was most ap- of model overfitting (Phillips 2010). parent in modeling regions where the variables had outlier Using more than just one evaluation statistic to evaluate values at the extreme high end of the distribution histogram. habitat model performance is highly recommended (Liu et The inclusion of the hinge and product features compen- al. 2005), so in addition to gain, we evaluated the area under sated for the omission of the quadratic feature, because in the receiver operating curve (AUC) statistic to determine combination, they can conform to a quadratic shape. We model accuracy and fit to the testing data (Fielding and also selected the “auto features” option, which allows Bell 1997). The AUC statistic is a measure of the model’s MaxEnt to further limit the subset of response features predictive accuracy, and it was originally developed for from those we selected above by retaining only those with evaluations using presence and absence data, producing an some effect. index value from 0.5 to 1 with values close to 0.5 indicat- Other techniques can be used to control overfitting ing poor discrimination and a value of 1 indicating perfect the data, such as reducing the number of parameters in the predictions. The AUC values can be interpreted similarly to model. To do this, MaxEnt provides a “regularization the traditional academic point system where values between

32 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

0.9 and 1.0 indicate an excellent model (A), 0.8 to 0.9 is list of variables was agreed to) to minimize the difference good (B), 0.7 to 0.8 is fair (C), 0.6 to 0.7 is poor (D), and between the regularized training gain and test gain, while AUC values between 0.5 and 0.6 represent failure (F), or maximizing the test AUC and CBI using the held-out test- models that don’t predict much better than a random guess. ing points. An example of this interpretation in the field of niche-based The final models used for reporting status and trends species distribution models can be found in Araújo et al. are the average summary statistic model outputs from these (2005) and Randin et al. (2006). In our situation, MaxEnt replicates. MaxEnt also produced other summary statistic uses 10,000 randomly selected background locations (map grids, such as the standard deviation for each cell within pixels) instead of true absence data, so it is not possible to the modeling region. We used these maps to calculate achieve an AUC value of 1.0 (Wiley et al. 2003). However, a 95-percent confidence interval (CI) for each cell and interpretation is similar, with higher AUCs indicating better produced upper and lower limit maps based on it. These model predictions (Phillips et al. 2006). Specific to our case, summary maps were used to generate histograms of the AUC values represent the percentage of times a spotted owl model predictions uncertainty for each model region and for nest site location would have a higher habitat suitability each bookend (app. C). The maps produced are also useful value than a randomly selected location from the modeling to see where within the modeling region the model predic- region. tions are less robust. Our third measure of model performance was the continuous Boyce index (CBI) as described by Hirzel et Environmental Variables al. (2006). This index and methodology is designed specifi- The environmental variables that influence the spotted cally for testing habitat suitability models produced from owl’s distribution in the Pacific Northwest have been well presence-only data. The index is based on the Spearman studied, and a wealth of information exists in the literature on important vegetation characteristics associated with owl rank correlation coefficient (Rs) that compares the ranks of modeled species occurrence with the area available to habitat use. As previously noted, we were restricted to only “binned” modeled prediction ranks (Boyce et al. 2002). A a few basic factors (i.e., tree diameter, canopy cover) for good model would predict an increasing ratio of the percent- the habitat modeling done in the 10-year monitoring cycle age of species occurrence to the percentage of the modeled (Davis and Lint 2005); however, the GNN map products landscape in each model bin as the bin values increase. An provided us a more extensive “menu” of forest vegetation variables to consider. Our initial selection of vegetation Rs of 1.0 indicates a strong positive correlation (Boyce et al. 2002). characteristics and environmental variables for habitat We produced 10 bootstrapped random replicates for modeling was based on three things: (1) habitat relation- each modeling region using 25 percent of the training data ship information in the literature or expert knowledge, (2) held out to test the model. We reviewed the jackknife graphs on-the-ground plot accuracies of the variable, and (3) cor- for mean test gain and AUC from these replicates, which are relations between the covariates. We chose not to use any produced by MaxEnt. These graphs illustrated the contribu- GNN structural or age variables that had plot correlations tion that each variable made to the overall model (Phillips less than 0.3 for an individual modeling region and <0.5, et al. 2006). Based on these graphs, we dropped variables averaged across all modeling regions. For species composi- that significantly increased mean test gain and AUC when tion variables, we chose not to include any variables that excluded. Once this decision was made, a final check for had kappas <0.2 for individual modeling regions or <0.3, model overfitting (see above) was conducted. This process averaged (as a species group) across all modeling regions. entailed increasing the regularization multiplier by incre- In cases where variables were highly correlated (Pearson ments of 0.5 from the default setting of 1.0 (once the final correlation >0.7) with each other we dropped the variable with the lower plot accuracy.

33 GENERAL TECHNICAL REPORT PNW-GTR-850

From our initial list of GNN variables, we dropped (a.k.a. geographic region) information from the population basal area of conifers ≥20 in diameter at breast height monitoring work (app. A in Anthony et al. 2006, Forsman (d.b.h.) because it was highly correlated with the mean stand et al. 2011) to combine some provinces and Environmental conifer diameter, stand height, and the diameter diversity Protection Agency level III ecoregions (Omernik 1987) index, but had the lowest plot accuracies. We also dropped to guide final delineations of modeling regions. Modeling the standard deviation of d.b.h. of all live trees for similar regions were only used for habitat modeling purposes, but reasons. We also did not include total canopy cover or stand we still report on habitat status and trend conditions within density index variables because both had high correlations the physiographic provinces to maintain consistency with with conifer cover, which had the highest plot accuracy of previous reports. all GNN variables. We considered, but did not use any GNN Within these modeling regions, our modeling back- variables for snags and down wood because of low plot ground (the area for which MaxEnt compares the combina- accuracies for those types of variables. tions of environmental variables that underlay owl locations We ended up with a consistent set of five variables that to the broader area that is available for use) was based on reflected forest structure and one forest age variable that we a “habitat-capable” mask that we generated specifically included in all of our modeling regions. The accuracy of the for habitat modeling purposes. The GNN environmental variables we used is shown in appendix A (table A-2), along data are modeled from detailed field plot data from forest- with Pearson correlations between covariates we selected capable areas only, and a non-forest-capable “mask” is for habitat modeling. We also developed five forest species provided by GNN using ancillary land class data from the composition variables (i.e., subalpine, pine, evergreen hard- Gap Analysis Program (GAP) and National Land Cover woods, oak woodlands, and redwoods) and included them Data (NLCD) data sets (Vogelmann et al. 2001). The GAP as appropriate for each modeling region (app. A, table A-2). data are based on multiseason satellite imagery (Landsat For instance, we did not include a subalpine variable in the ETM+) from 1999 to 2001 used in conjunction with other California Coast Range modeling region, because none ex- data sets (i.e., elevation, landform, aspect, etc.) to model the ists in that area. Likewise, we did not include the redwood distribution of ecological systems (Comer et al. 2003) and variable in the western Washington/Olympic Peninsula land cover classes at a 1-ha (2.47-ac) resolution. However, modeling region. The final list of variables used in each upon review, the GNN mask included inconsistent masking modeling region is provided in appendix C. of urban areas and roads, and also did not mask out areas that we felt were not capable of developing into habitat Modeling Regions (i.e., subalpine parklands and steppes). Therefore, we Based on recommendations from the 10-year report (Davis used the “unmasked” GNN data set and applied our own and Lint 2005), we developed habitat modeling regions that customized mask specific to our purposes. The mask we removed some administrative boundaries (i.e., state lines) developed included the use of the “impervious layer” from and framed areas based more on ecological rather than so- NLCD (Herold et al. 2003) to consistently exclude areas that ciopolitical divisions. Our modeling regions were modified have been converted into non-habitat-capable conditions versions of the standard physiographic provinces developed (i.e., urbanized areas, major roads, etc.) and refined the in FEMAT (1993) and used for reporting monitoring results developed open space designations. We then modified the (fig. 3-5). Our intent was not to further split the existing existing GNN mask classes to exclude a few additional land delineations into smaller areas, but to combine the existing classes or ecological systems that we felt were not habitat delineations based on two things: (1) ecological similarities capable. Isolated areas less than 2/3 ac (pixel map noise) of between physiographic provinces and (2) occurrence and both mask and nonmask were removed. The intent of our distribution of spotted owl location data being used for mask was to frame our modeling area such that it contained model training and testing. We used the ecological region lands capable of producing closed-canopy forests that could

34 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3-5—Modeling regions used for modeling northern spotted owl habitat.

35 GENERAL TECHNICAL REPORT PNW-GTR-850

be potentially suitable for spotted owl nesting, roosting, expected (P/E) ratio curves” (Hirzel et al. 2006). The con- foraging, or dispersal; however, we suspect that this mask tinuous P/E curves provide three indications of a model’s contains areas, especially in the higher elevations, that performance (Hirzel et al. 2006): might not actually be capable of developing habitat under 1. For replicated model runs that use held-out testing the current climate. methods (i.e., bootstrap or jackknife), the variance Habitat Map Development and Evaluation along the curve gives information about the model’s robustness along its range of probabilities. Smaller The MaxEnt model output is a logistic probability estimate variances indicate more reliable prediction points. of a site’s suitability for species presence based on environ- Large variances indicate the range of prediction values mental conditions from where the species are found, and that are the least robust. This information allows a their differences from the surrounding background envi- better understanding of the model’s strengths and ronmental conditions within the modeling region (Phillips weaknesses. and Dudík 2008). In our case, the environmental predictor variables used were based on stand-level structural and 2. The shape of the curve provides clues about the model’s forest-type species conditions associated with nesting and predictive power. The Spearman rank correlation roosting use by spotted owls. Therefore, our raw model out- coefficient (Rs) is used to help us judge the shape of put maps show a scale of nesting/roosting suitability (from the curve and the model’s performance. For fluctuating low to high) for forested stands based on the stand structure curves, each time the curve dips as the ranks increase, and species composition conditions described by the GNN Rs decreases. A higher Rs indicates a consistently in- data. The mapped logistic probability values will be higher creasing larger proportion of species presence (versus where these stand-level conditions are more similar to the available) being predicted as the model prediction conditions observed where we have documented nests and output increases. This is indicative of a good model; territorial pair centers (i.e., the training data). A mapped however, note that one can get the same Rs for many logistic probability of 0.5 represents the “average” condition different-shaped curves (i.e., linear, exponential, and where the species occurred (Phillips 2008). sigmoid), and curves with flatter slopes can have the Our charge is to develop habitat maps that work well same ranks as curves with steep slopes. According to and to then measure and report on amounts and distribution Hirzel et al. (2006), a perfect model would have a linear of habitat. The latter requires that we select a threshold P/E curve that monotonically increases as probability from the probability values described above to represent increases because a perfectly straight line allows for an “suitable” owl nesting/roosting habitat for summarization infinite number of classes along the scale of probability purposes. In the 10-year report, we used the area-adjusted (i.e., “resolution”). A wavy line lowers the resolution frequency (AAF) curves (Boyce et al. 2002) associated because classifying the line depends on these changes with the habitat suitability output from BioMapper to in the shape and slopes. For exponential models evaluate our habitat models. These curves are among the (like MaxEnt), an exponentially increasing curve is few diagnostic measures designed specifically for measur- indicative of a good model. ing the accuracy of habitat models based on presence-only 3. The maximum y-axis value reached by the P/E curve data (Hirzel et al. 2006). But in addition to model evalua- reflects how much the model differs from chance tion, these curves also provide information that can be used expectation, or deviation from randomness. This score to reclassify habitat models into discrete habitat classes reflects the model’s ability to differentiate the species (Hirzel et al. 2006). To conform to the new terminology, we niche characteristics from those of the modeled region. now refer to AAF curves as “continuous predicted versus Caution is needed because this maximum value is sensitive to the species niche breadth within the context

36 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

of the modeled region. In other words: Does the species via forest succession over the next few decades. The only just use a small percentage of what is available in the difference between the two reports is that the classes in the modeled region or is its habitat use more generalized 15-year report are based on commonly used thresholds and within the modeling region? If there is abundant habitat have more biological meaning. These habitat classes are available in the modeling region that is being used by defined as follows: the species, the model will usually produce a flatter • Unsuitable—MaxEnt logistic output from zero to curve with lower P/E values. Also, the selection and the mean value between zero and the P/E = 1 thresh- resolution of the environmental variables used for old. This habitat class represents the lowest suitabil- modeling can influence the maximum P/E value. ity class and owls will normally avoid using it for Once the habitat model evaluation process has been nesting and roosting. completed, the P/E curve provides a method for classifica- • Marginal—MaxEnt logistic output from the mean tion of a model into discrete habitat classes (Hirzel et al. value between zero and the P/E = 1 threshold to the 2006). The point along the model prediction axis (x-axis) P/E = 1 threshold. This habitat class represents a where the curve crosses P/E = 1 along the y-axis (fig. 3-6) is condition approaching what owls will nest and roost the threshold where the model predicted species occurrence in. Occasionally, these habitat characteristics are is higher than would be expected if there were no selection associated with nesting and roosting owls; however, (i.e., habitat use was random). This threshold is often used this could be due to occurrence of legacy habitat fea- to classify habitat models into binary maps, where logistic tures such as large trees, extreme rarity of suitable probability values greater than the P/E = 1 threshold repre- nesting/roosting habitat, or perhaps interspecific sent “suitable” habitat (Hirzel et al. 2006). We also note that competition with barred owls. in our case, the P/E = 1 threshold was similar to the “maxi- • Suitable—MaxEnt logistic output from the P/E = 1 mum specificity and sensitivity threshold”3 (Phillips and threshold to 0.5. A MaxEnt logistic output value of Dudík 2008) for all model regions. We provide these and 0.5 represents the “average” environmental condi- additional thresholds, that are commonly used, in appendix tion associated with the owl training data. This C. We also note that the 10-percentile threshold (app. C) is habitat class represents habitat conditions where the equivalent to where we reported that 90 percent of the owl probability of owl presence is higher than expected training data occurred in the 10-year report habitat models by random chance and up to average conditions as- (see fig. 3-11 and table 3-4 in the 10-year report). sociated with nesting and roosting. We further divided the continuous scale of probability • Highly suitable—MaxEnt logistic output from 0.5 of occurrence from our habitat models into four habitat to the highest output from the habitat model. This classes that represent from the least to the most suitable habitat class represents the most suitable, or “above habitat conditions (fig. 3-6). This was done to produce histo- average,” conditions used by nesting and roosting grams (app. F) similar to the five-class histograms used to territorial owl pairs. profile the continuum of habitat suitability in the 10-year In some of the modeled regions, the 10-percentile report (Davis and Lint 2005). As in the 10-year report, threshold occurs within the “marginal” habitat class indicat- tracking the changes in these habitat profiles (app. F) is ing some owl nesting/roosting use of younger, mid-aged expected to provide useful information for visualizing stands as noted by Thomas et al. (1990) who stated that as where habitat may be recruited (first two habitat classes) forests develop along the continuum from young to old, they 3 Minimizes omission (false absence predictions) and commission gradually become more suitable for spotted owl nesting/ (false presence predictions) errors. roosting. To show this continuum of conditions, and to help interpret what these habitat classes represent on the ground,

37 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 3-6—The predicted versus expected ratio curve (modified from fig. 6 in Hirzel et al. 2006).

we also provide average stand structure and age attributes Although late-successional and old-growth forests are often (table 3-1). It appears that the lowest class of habitat includes equated with spotted owl habitat, they are not always the early- to mid-successional forests and the highest suitability same. As noted by Thomas et al. (1990), the redwood zone class includes the oldest and most structurally complex for- in northwestern California is unique in terms of owl habitat ests (table 3-1). However, we stress that these simple combi- development. In that portion of the owl’s range, the structur- nations of forest attributes do not fully describe habitat, and al conditions that constitute nesting/roosting habitat develop it is the complex interaction between them that does. quicker, with suitable conditions occurring in 40 to 60 years on some sites and superior conditions in 80 to 100 years. Nesting/Roosting Habitat Habitat development is not a mechanistic process, and The importance of mature or late-successional forests for there is considerable variability in predictions of habitat nesting, roosting, and foraging of owls in the Pacific North- (Courtney et al. 2004). As can be seen in table 3-1 and west is clear (see reviews in Thomas et al. 1990), with nu- appendix C, the transition from unsuitable to suitable con- merous studies documenting both selection of these habitats ditions is more complex than a simple increase in a stand’s by owls (Carey et al. 1990, Forsman et al. 1984, Glenn et average tree diameter and canopy closure. In addition, spe- al. 2004, Gutiérrez et al. 1984, Hamer et al. 1989) and now cies composition is also important; for instance, late-succes- more recent research linking greater amounts of older forest sional/old-growth ponderosa pine forests do not function as in owl territories to owl fitness (i.e., increased survival and/ nesting/roosting habitat, nor do older subalpine forests. or reproductive success) (Dugger et al. 2005, Franklin et We consider our “suitable” and “highly suitable” al. 2000, Olson et al. 2004). High-quality owl habitat was habitat classes, as described above, as nesting/roosting described by Thomas et al. (1990) and generally includes habitat. It is important to emphasize that our maps are not older, multilayered, structurally complex forests character- attempting to predict owl occupancy or other demographics ized by large-diameter trees, high amounts of canopy cover, across the landscape, but rather describe stand-level habitat numerous large snags, and lots of downed wood and debris. characteristics that are associated with owl pair use that

38 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

±89 ±78 ±54 ±58 ±45 ±65 ±83 ±80 ±102 ±57 ±76 ±65 ±47 ±20 ±21 ±26 ±49 ±60 ±45 ±47 ±32 ±22 ±74 ±88 ears verage Y 40 73 87 88 23 46 74 50 82 52 76 35 47 57 78 137 106 128 137 123 151 205 185 A 111 stand age

±25 ±26 ±28 ±31 ±30 ±26 ±32 ±25 ±20 ±20 ±23 ±26 ±22 ±26 ±24 ±31 ±21 ±24 ±25 ±27 ±16 ±20 ±31 ±30 4 5 Feet 42 76 94 45 57 70 85 36 74 37 68 91 33 50 66 95 38 48 63 84 verage 11 11 106 143 A stand height

±2 ±2 ±1 ±1 ±2 ±1 ±1 ±1 ±2 ±1 ±1 ±1 ±2 ±1 ±1 ±1 ±2 ±2 ±2 ±1 ±2 ±2 ±2 ±1 2 5 6 7 3 4 5 6 2 6 7 2 5 6 7 2 4 6 3 5 5 6 4 7 Index diversity Diameter

±8 ±8 ±8 ±7 ±3 ±6 ±8 ±3 ±3 ±5 ±7 ±1 ±6 ±2 ±5 ±8 ±3 ±5 ±7 ±2 ±2 ±7 ±8 ±2 7 3 9 1 2 3 6 0 7 2 7 1 1 3 5 7 1 1 1 1 11 15 19 16 rees/acre T Large conifers (≥30-in d.b.h.)

d.b.h. ±7 ±9 ±9 ±7 ±6 ±6 ±7 ±9 ±8 ±8 ±6 ±7 ±6 ±9 ±9 ±9 ±11 ±10 ±10 ±16 ±14 ± 10 ±12 ±9 8 verage 11 22 17 24 30 13 16 17 20 19 26 36 17 29 29 18 11 11 25 Inches 13 19 24 24 A conifer

±8 ±5 ±9 ±8 ±11 ±29 ±12 ±31 ±15 ±10 ±31 ±19 ±15 ±10 ±26 ±15 ±22 ±20 ±18 ±17 ±21 ±20 ±20 ±15 cent 85 89 81 82 76 42 79 48 60 75 37 61 65 70 38 70 24 51 60 65 16 44 64 78 cover Per Conifer

0–6 0–9 0–9 0–1 7–25 0–15 0–12 26–50 51–86 12–35 36–50 51–93 10–28 29–50 51–91 10–30 31–50 51–88 16–37 38–50 51–86 13–35 36–50 51–86 Habitat suitability

ginal ginal ginal ginal ginal ginal class Habitat Unsuitable Mar Suitable Highly suitable Unsuitable Mar Suitable Highly suitable Unsuitable Mar Suitable Highly suitable Unsuitable Mar Suitable Highly suitable Unsuitable Mar Suitable Highly suitable Unsuitable Mar Suitable Highly suitable ± standard deviation) habitat variable values (gradient nesting/roosting nearest for neighbor) habitat classes

Coast and Cascades Eastern Cascades Coast Range California Cascades California Klamaths Coast d.b.h. = diameter at breast height. Table 3-1–Average ( 3-1–Average Table in each modeling region Model region Washington Washington Oregon Oregon Oregon California a Note: This table is intended to provide a general sense stand of structure variable gradients from unsuitable to highly suitable.

39 GENERAL TECHNICAL REPORT PNW-GTR-850

approximates a species’ realized niche within a specific We did not use presence locations and MaxEnt to environmental space (Phillips et al. 2006). model dispersal habitat. Instead we developed dispersal habitat maps for both bookend periods using simple GIS Dispersal Habitat queries of our GNN variables for conifer d.b.h. ≥11 in and Dispersal habitat is used by juvenile owls moving away conifer cover ≥40 percent, similar to what was done in the from natal areas or by subadults and adults moving between 10-year report (Davis and Lint 2005). We also included both territories (Forsman et al. 2002). Spotted owls are capable suitable habitat classes from our nesting/roosting habitat of dispersing long distances, and gene flow from one por- models, because owls obviously disperse through nesting/ tion of the range to another can occur in a few generations roosting habitat. We then analyzed the status and trend of (Forsman et al. 2002). The network of large reserves this habitat within federal reserved and nonreserved LUAs, established under the Plan appeared suitable for maintain- as well as nonfederal lands. ing interconnected populations of spotted owls (Lint et To detect changes in amounts of dispersal habitat that al. 2005); however, concern remained for disjunct small might affect owl movement across the landscape, we con- populations that are isolated by large nonforested areas or ducted a landscape-scale analysis using a spatial frame- expanses of young managed forests (Forsman et al. 2002). work based on Forsman et al. (2002). Only 8.7 percent of Thomas et al. (1990) predicted that much of the forested dispersing individuals moved more than 31 linear mi and area between owl conservation areas would be suitable for only “large expanses” of nonforested or younger forested passage by dispersing spotted owls as long as at least 50 areas appear to pose significant barriers to this movement percent of the landscape was forested with conifer stands (Forsman et al. 2002). We used this distance to define the with an average d.b.h. of ≥11 in with at least 40 percent radius (15.5 mi) for a circular analysis window within canopy closure. This definition of a dispersal-capable which we quantified the percentage of dispersal habitat for landscape became known as the “50-11-40 rule” (Thomas et both bookend periods and included all landownerships. al. 1990) and was based on information of habitat conditions This distance is also comparable to the root-mean-square for dispersing juvenile owls (Miller 1989). Older forest dispersal distance (a measure of gene flow) estimated by habitat is more frequently used for natal dispersal, but Barrowclough et al. (2005). We then overlaid linear owl closed-canopy (>60 percent cover) younger forests are also dispersal paths from the 10-year report (Lint et al. 2005) used, whereas younger open-canopied (<40 percent cover) on the baseline version to measure underlying percentages forests are generally avoided (Miller et al. 1997). Dispersal of dispersal habitat in the landscape through which they distance is also negatively associated with the amount of dispersed (fig. 3-7). The mean percentage of dispersal clearcut forest in the landscape, and large urban and agricul- habitat for both juvenile and nonjuvenile owls was 55 tural areas appear to be barriers to dispersal (Forsman et al. percent. We combined results across age classes and used 2002, Miller et al. 1997). Spotted owls use a wide variety of the 10-percentile value (40 percent) from all owl dispersal forest habitats for dispersal and will traverse very frag- paths as a threshold to create binary maps from the roving mented landscapes (Forsman et al. 2002), but little informa- window analysis maps. Thus, the binary maps show where tion exists on how the amount or fragmentation of habitat there appears to be enough dispersal habitat at the landscape influences dispersal. The results of the latest meta-analysis scale (≥40 percent within a 15.5-mi radius) to accommodate suggest that recruitment into the territorial breeding popula- 90 percent of known owl movements. We call this footprint tion may depend on the presence of sufficient amounts of the “dispersal-capable landscape” and used it to identify high-quality dispersal habitat, enough to ensure survival of potential disconnects or bottlenecks for owl movement be- dispersing owls until they recruit into the territorial popula- tween large block reserves. We also identified areas across tion (Forsman et al. 2011). the range of the owl where the footprint shrank or expanded between our bookends.

40 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3.7—Juvenile and nonjuvenile dispersal straight-line paths from Lint et al. (2005) in relation to the amount of dispersal habitat within a 15.5-mi radius that was based on information from Forsman et al. (2002) where only 8.7 percent of dispersing individuals moved more than 31 linear mi.

Habitat Fragmentation recent research on this subject shows that habitat fragmenta- Although large blocks of contiguous, high-quality habitat tion can affect occupancy and other demographic factors, provide the best configuration for long-term persistence and may result in isolated populations and interruption of of owl populations (Thomas et al. 1990), smaller blocks or gene flow (Courtney et al. 2004). patches of owl habitat can also be important as dispersal In a general sense, habitat can be divided into two habitat (Forsman et al. 2002). These smaller patches help to broad landscape morphological categories: (1) core habitat, maintain connectivity between the larger blocks of habitat which occurs only in larger habitat patches and is some that will eventually develop in the reserve system designed distance away from the patch edge (sometimes referred to as under the Plan. At the time of the owl’s listing, habitat frag- “interior habitat”); and (2) edge habitat, which occurs along mentation was believed to be a stressor for spotted owls the margins of larger habitat patches surrounding the core because it is associated with habitat loss, and was also habitat or occurs in patches that are too small to contain thought to improve habitat conditions for spotted owl preda- core habitat. tors, such as the great horned owl (Bubo virginianus) (Carey It is not clear how habitat fragmentation affects owl de- et al. 1992). There is no clear evidence of indirect effects mographics; however, survival and reproduction are higher of fragmentation through predation, but it remains as a on owl territories with more old-forest habitat centered on possible threat (Courtney et al. 2004). A compilation of the the nest tree or activity center (Dugger et al. 2005, Franklin

41 GENERAL TECHNICAL REPORT PNW-GTR-850

et al. 2000, Olson et al. 2004). Edge habitat also appears to be important to spotted owls in some portions of their range, probably as a source of prey (Franklin et al. 2000, Olson et al. 2004; but see exception in Dugger et al. 2005). Here we define core habitat as the internal portion of a stand of nesting/roosting habitat that is farther than 100 m from the stand edge. Edge habitat is defined as all noncore nesting/roosting habitat and is always adjacent to nonhabitat. We do, however, distinguish between two types of edge habitat: (1) core-edge habitat, which is the amount of nesting/roosting habitat adjacent to and surrounding core patches (i.e., the edges of large habitat patches), and (2) all other edge habitat that is not directly adjacent to core habitat (i.e., small, isolated habitat patches). In juxtaposition, core and core-edge habitat reflect more contiguous habitat blocks, whereas large amounts of non-core-edge habitat occur in landscapes that are highly fragmented, with patch sizes too small to contain core habitat. We used GUIDOS v1.3 (Soille and Vogt 2009) to conduct a morphological spatial pattern analysis (MSPA) on 100-m (2.47-ac)-resolution binary raster (grid) maps of nesting/roosting habitat for both 1994/96 and 2006/07 to assess status and trend in habitat configurations. GUIDOS was specifically developed for analysis of forest spatial patterns extracted from satellite images (Soille and Vogt 2009, Vogt et al. 2007). It produces simple-to-interpret maps of core and edge patterns from binary raster maps, and the outputs are pixels with specific core or edge classifications (fig. 3-8). From this product, we conducted an area analysis that quantifies the area represented by both types of pixels (“core” or “edge”); thus, in our analysis, edge is not quantified as a perimeter. Specifically, edge habitat only occurs within 1 pixel width, or 100 m (328 ft), from a nonhabitat pixel, and, therefore, core habitat pixels are greater than 328 ft from nonhabitat pixels. This distance is similar to that used by Franklin et al. (2000) and Zabel et al. (2003) to define their core habitat. Using 100-m (2.47-ac)-resolution maps requires a patch of contiguous habitat to be greater than 22 ac before it can contain core habitat. Therefore, the combination of core plus core-edge pixels shows patterns of habitat patches that are at least that large. All patches of Figure 3-8—Example of the morphological spatial pattern nesting/roosting habitat smaller than that are essentially analysis (MSPA) on binary maps of nesting-roosting habitat.

42 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

edge habitat. We report on the status and trend of core Olympic (Rs = 0.93), and, surprisingly, the Klamath Moun- habitat and changes in the percentage of the sum of [core] + tain modeling region (Rs = 0.92). The poorest model projec- [core-edge] habitat to all nesting/roosting habitat. This per- tions occurred within the Washington Cascades modeling centage can serve as an index of landscape habitat fragmen- region (Rs = 0.74) and California Coast Range (Rs = 0.63). tation, as the higher the percentage, the more contiguous the During this testing process, we noted interesting differences habitat is within the landscape and the lower the percentage, between the average habitat suitability values where spotted the more fragmented the habitat (fig. 3-8). owls in the demographic study areas occurred in 1994/96 compared to where they occurred in 2006/07 (fig. 3-10). Results We observed consistently lower than average habitat suit- Habitat Suitability Modeling ability values in 2006/07 compared to 1994/96; however, Our final habitat models and map products (fig. 3-9) repre- 95-percent CIs overlap between periods. We speculate sent the mean from 10 bootstrapped replicates. We decided that spotted owls might be using lower quality habitat in to use the means as our product, because the P/E curves that 2006/07 because they are being displaced from higher are generated for the means provide the users with valuable quality habitats by barred owls, whose density has increased information on how to interpret the model (see the “Habitat steadily since the late 1990s (Forsman et al. 2011). The Map Development and Validation” section and fig. 3-6). We potential for displacement of spotted owls by barred owls also provide the summary statistic maps (i.e., 95-percent CI) in the current bookend is the reason we trained our models to supplement this interpretation with area-specific informa- using the 1994/96 spotted owl locations. However, given the tion as discussed in the methods section. aforementioned issues on model projection [extrapolation], Performance of bookend 1 (1994/96) models was fair these results, based on our bookend 2 models, should be to good (i.e., C+ to B+ grades) with AUCs ranging from interpreted with some caution. 0.78 to 0.88 and Spearman rank correlation coefficients Nesting/Roosting Habitat >0.9 (P < 0.001) (app. C). Our lowest performing models 4 occurred in the Oregon and California Klamath and We estimate a rangewide gross loss of about 298,600 ac of California Coast modeling regions, and our best models spotted owl nesting/roosting habitat on federal lands (app. were in the Washington Coast and Cascades and Oregon D). This amounts to about 3.4 percent of what was pres- Coast modeling regions. We suspect this is because of the ent in 1994/96 (bookend 1). Most of the loss (79 percent) rich vegetative diversity in that area that (1) confounds occurred within the reserved allocations, which amounted remotely sensed data development and (2) produces a more to about 3.7 percent of the reserved areas under the Plan, complex “definition” of habitat because of the complex whereas nonreserved allocations experienced a 2.7 percent variable interactions. Regardless of the reason, the model loss of habitat. Wildfires remain the primary cause of AUCs for these regions were 0.78 and 0.81, respectively, and habitat loss, accounting for about 90 percent of the loss in therefore provide useful information (Swets 1988). reserved allocations (203,900 ac), and about half of the loss Our projected models (bookend 2, 2006/07) were tested in nonreserved allocations (32,600 ac). Timber harvesting using the 2006/07 owl location data sets not used for model accounts for about 45 percent of the loss in nonreserved training. Spearman ranks based on the continuous Boyce allocations (37,400 ac) and 7 percent within reserved index (Hirzel et al. 2006) ranged from 0.63 to 0.98. The allocations (16,600 ac), and insects and disease outbreaks best model projection [extrapolation] occurred in the account for about 3 percent of the loss in all allocations Oregon Cascades modeling region, followed in order by the (fig. 3-11). Relative to the baseline maps, and based on

Oregon Coast Range (Rs = 0.95), western Washington and 4 Acres are rounded up to the nearest 100 ac.

43 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 3-9—Northern spotted owl habitat suitability map showing the spatial distribution of nesting/roosting habitat as of 2006 (in Oregon and Washington) and 2007 (in California).

44 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3-10—Observed differences in average modeled habitat suitability for spotted owl pair locations within demographic study areas between 1994/96 and 2006/07.

Figure 3-11—Causes of nesting/roosting habitat loss on federally administered lands.

45 GENERAL TECHNICAL REPORT PNW-GTR-850

LandTrendr change-detection data, the physiographic habitat loss in the Oregon western Cascades occurred in the province that experienced the greatest loss of habitat was southern half of that province. the Oregon Klamath province because of the large Biscuit Because wildfires appear to be the number one cause Fire that occurred in 2002. In general, the Klamath and of habitat loss, we conducted a more indepth analysis of the eastern Cascades physiographic provinces experienced the 20 largest wildfires that occurred within the owl’s range largest percentage losses of habitat related to wildfires (fig. between 1996 and 2006 (years with satellite data across 3-12); however, in terms of absolute acreage of habitat lost, the range). Table 3-2 lists these fires in descending order the Oregon Klamath ranked first (93,600 ac), California of estimated acres of owl habitat lost. Overall, these 20 Klamath ranked second (71,600 ac), and the Oregon western fires accounted for almost 200,000 ac of habitat lost. The Cascades ranked third (28,900 ac) (app. D). Most of the percentage of owl habitat lost within their fire perimeters differed among the east and west Cascades (Washington and Oregon) and the Klamath Mountains (Oregon and California) physiographic provinces (fig. 3-13). The percentage lost per fire in the Klamath Mountains and the west Cascade provinces were not significantly different (overlapping 90-percent CIs); however, percentage of habitat loss per fire was notably higher in the eastern Cascades. However, in terms of the amount of nesting/roosting habitat burned by these 20 fires, the vast majority of acres lost occurred in the Klamath Mountains (143,000 ac), followed by the east side of the Cascades (36,000 ac) and the western Cascades (20,000 ac). Based on Climate, Ecosystem, and Fire Applications (CEFA) program data (Brown et al. 2002) and wildfire perimeter data (MTBS and GeoMac), wildfires burned an estimated 2.6 million ac within the owl’s range between 1994 and 2007, which frames our analysis period. From our observations, it is clear that wildfires do not remove all owl nesting/roosting habitats within their perimeters. Fires of low to moderate severity can alter this habitat, but do not necessarily result in its loss. The commonly used term to define this effect is “habitat degradation.” We estimated owl habitat degradation, as the number of acres that changed from the “highly suitable” to the “suitable” habitat class between our bookends (1994/96 to 2006/07). For habitat degradation, our analysis showed the reverse trend from what we observed for habitat loss (fig. 3-14). These results suggest that wildfires in the east Cascades have been more destructive (higher amount of habitat loss, lower amount of degradation) and that wildfires in the west Cascades and Figure 3-12—Nesting/roosting habitat trends (based on the LandTredr analysis) from 1994/96 to 2006/07 by physiographic Klamath Mountains were less severe, producing a mosaic of province for reserved and nonreserved federal lands. fire effects indicative of a moderate-severity regime.

46 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Table 3-2—Estimated fire effects on northern spotted owl nesting/roosting habitat from the 20 largest wildfires between 1996 and 2006 Broad Habitat in physiographic wildfire Fire name Year province perimeter Habitat lost Habitat degraded Acres Acres % Acres %

Biscuit Fire 2002 Klamath Mountains 226,230 93,730 41 12,019 5 Megram Fire 1999 Klamath Mountains 76,337 27,520 36 4,589 6 B&B Complex 2003 East Cascades 26,269 16,403 62 907 3 Bake-oven Fire 2006 Klamath Mountains 23,946 8,873 37 581 2 Boulder Fire 2002 West Cascades 34,059 8,460 25 2,074 6 Davis Fire 2003 East Cascades 8,050 6,943 86 5 0 Pigeon Fire 2006 Klamath Mountains 13,896 5,634 41 327 2 Rex Complex 2001 East Cascades 8,548 4,750 56 278 3 Timbered Rock 2002 West Cascades 10,216 4,539 44 569 6 Spring Fire 1996 West Cascades 13,504 3,931 29 858 6 Deep Harbor Fire 2004 East Cascades 5,761 3,930 68 64 1 Hancock Fire 2006 Klamath Mountains 12,712 3,132 25 336 3 Apple Fire 2 2002 West Cascades 12,227 2,810 23 928 8 Fischer Fire 2004 East Cascades 4,479 2,340 52 34 1 Fork Fire 1996 Klamath Mountains 2,962 2,113 71 14 0 Needles Fire 2003 East Cascades 1,946 874 45 1 0 Trough Fire 2001 Klamath Mountains 1,851 798 43 4 0 Hunter Fire 2006 Klamath Mountains 2,236 789 35 40 2 Deer Point Fire 2002 East Cascades 505 380 75 0 0 Tatoosh Complex 2006 East Cascades 666 378 57 0 0

Figure 3-13—Provincial differences in nesting/roosting habitat losses from the fires in table 3-2.

47 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 3-14—Provincial differences in nesting/roosting habitat degradation from the fires in table 3-2.

For this report, we were cautious in our use of the the satellite signals associated with the structural changes new GNN/LandTrendr data for measuring gains in nesting/ as forests progress to maturity and old age. Moreover, roosting habitat. Although the products we used for our small-scale forest canopy gap dynamics cannot be directly analysis (and other remote sensing approaches) have observed with the sensors used in our analysis (Frolking demonstrated their ability to detect both losses and gains in et al. 2009). Rather, all structural changes associated with forest cover (Coops et al. 2010; Hais et al. 2009; Kennedy et maturing forests often must be inferred from changes in al. 2007, 2010; Staus et al. 2002), the underlying measure- the spectral signal caused by proxy effects, such as within- ments from passive optical satellite sensors (i.e., those that canopy shadowing. Therefore, it is difficult to distinguish at take pictures of sunlight reflected from the Earth’s surface) a given location small changes in forest structure (and any place constraints on the subtlety of coniferous forest change associated variables, such as age) from background random that can be reliably captured over a short period (i.e., 10 to noise caused by differing sun angles, atmospheric effects, 12 years). Disturbances that result in substantial removal and phenological differences, particularly when the interval or reduction of vegetation cover (usually abrupt changes) of change is short (as for the 10- to 12-year period here) are easier to discern during change-detection than minor (Kennedy 2010). disturbances that cause more subtle change, or gradual dis- During our analysis, we conducted visual and GIS turbances that occur over a longer period, such as insect and examinations of our nesting/roosting habitat maps and vari- disease disturbances. Vegetation recovery can also be more able maps using aerial imagery and noted that commercial difficult to detect (depending on the type of vegetation and thinning of young plantations created suspicious changes timeframe), as it usually recovers gradually over a longer in some of our habitat modeling variables in the bookend period. Increases in tree bole diameter and forest canopy 2006/07 data set. For instance, in some modeling regions, cover happen at a faster rate in younger coniferous forests stand age increased by three to five decades, or density of than in older forests, however, and the satellite-measured large conifer (>30 in d.b.h) increased by as much as three to signal changes faster as well. Within the 10- to 12-year pe- four trees per ac, which is not likely within the timeframe of riod of this investigation, mapping of such changes in early our analysis. We suspect that canopy shadowing increased successional, pre-canopy-closure conditions are relatively owing to the thinning and may have caused some stands to robust (Kennedy 2010). Much more subtle, however, are

48 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

appear older than they actually were, thus making them ap- Dispersal Habitat pear as habitat when the modeling results from the 1994/96 Although we were cautious in our interpretation of gains in bookend were extrapolated (projected) to the 2006/07 nesting/roosting habitat, we feel that the GNN/LandTrendr bookend. data were better suited for detecting gains in younger for- We also conducted an analysis of the regional inventory ests (as described above), such as dispersal habitat, plus we plot data, similar to what was done in the 10-year report did not develop and then project (extrapolate) a dispersal (Davis and Lint 2005), to determine if there were significant habitat model from one period to another as we did for nest- gains of forest stand conditions that were similar to spotted ing/roosting habitat (i.e., no model transferability issues). owl nesting/roosting habitat (see table 3-5, page 47 in Davis Examination of the bookend changes in the two variables and Lint 2005). The results of this analysis did not show any that were used to define dispersal habitat (d.b.h. and conifer significant gains in “habitat classes” between the initial plot cover), and visual examination of the dispersal habitat maps measurement and the remeasurement data, which roughly overlaid on high-resolution color aerial imagery showed covers the same periods as our bookend models (app. H). In realistic changes that one might expect in a 10-to 12-year addition, the net changes between the bookend models were timeframe. well within the 95-percent CIs between periods; therefore, Rangewide, we report an estimated gross loss of about it is not possible to state with certainty that we observed 417,000 ac of dispersal habitat, most (82 percent) from wild- “real” net changes in nesting/roosting habitat between our fire (341,800 ac). The causes for dispersal habitat loss were bookend maps (app. C). For these reasons (plus the need for similar to those for nesting/roosting habitat losses, with caution when transferring or projecting models discussed wildfire being the main cause in reserved allocations and earlier), we focused on habitat losses, which are more accu- about half of the loss in nonreserved allocations (fig. 3-15). rately detected with current technologies and were verified Timber harvesting accounts for the other half of the loss by LandTrendr change-detection data. For the next round in nonreserved allocations, and insects and disease account of monitoring (20-year report), we hope to use LandTrendr for a small percentage of loss in all allocations (fig. 3-15). for verification of both nesting/roosting habitat losses and However, these losses were offset by a 1.26-million-ac gains. gross gain in dispersal habitat on federal land from forest

Figure 3-15—Causes of dispersal habitat loss on federally administered lands.

49 GENERAL TECHNICAL REPORT PNW-GTR-850

succession, resulting in a 5.2-percent overall net gain of dispersal habitat coverage across the owl’s range (app. E). In general, the gains in dispersal habitat were higher in federal nonreserved allocations than in reserved allocations. Only the Oregon Klamath experienced a net decrease in the amount of dispersal habitat (-2.6 percent) owing to the large Biscuit Fire, which removed more dispersal habitat than was recruited for this period (app. E). The biggest net gain (13.1 percent) in federal dispersal habitat occurred in the Oregon Coast Range, which has some of the most productive forests in the owl’s range. An example of this recruitment is clearly seen in the maps from 1996 and 2006 for the large Oxbow Fire of 1966 (fig. 3-16). In 1996, this area was forested with stands just about 30 years of age. Based on tree diameter growth data for fully stocked, site class 1, Douglas-fir (Pseudotsuga menziesii [Mirb.] Franco) forests, stands of this age have an average d.b.h. of 9 in and can put on 3 in of diameter growth in one decade (McCardle et al. 1961), thus crossing the threshold from nondispersal to dispersal habitat in a relatively short timeframe. However, not all sources of gain for dispersal habitat come from forest succession. Sometimes disturbances, such as a moderate-severity wildfire, can alter (i.e., opening up the canopy) suitable nesting/roosting habitat, making it unsuitable for nesting and roosting, but still suitable enough for owl dispersal (see table 3-1). At the landscape scale, we detected a 5-percent gross loss of dispersal-capable landscape, mostly around the periphery of the federal forests. We suspect this may be due to regeneration timber harvesting occurring in dispersal habitats on nonfederal lands that border federal lands. Large wildfires on federal lands played a role in this decrease in the eastern Cascade provinces and the Oregon Klamath Mountain province. We also detected a 4-percent gross gain in dispersal-capable landscapes along the periphery of some federal forests caused by forest succession in younger forests, resulting in an overall net decrease of 1 percent in dispersal-capable landscape area (fig. 3-17). The most noticeable change in dispersal-capable landscapes, that we detected, occurred in the northeastern portion of the Washington eastern Cascades; the losses of dispersal-capable landscape caused by large wildfires in Figure 3-16—Recruitment of dispersal habitat in the Oxbow Fire (1966) in the Oregon Coast Range. 50 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 3-17—Changes in dispersal-capable landscapes across the owl’s range.

51 GENERAL TECHNICAL REPORT PNW-GTR-850

that area may have isolated some of the large LSRs established at the Plan’s implementation (fig. 3-17). This may also be the case for the LSR just to the east of the B&B Fire where it appears that there has been a 3- to 6-mi contraction of dispersal-capable landscape in that area. Overall, the large reserved network still appears to be well connected, with the exception of three areas. Of primary concern are the federal reserved lands on the Olympic Peninsula, which are separated from the Cascades by about 75 mi of landscape with poor dispersal conditions (fig. 3-17). These federal lands are also separated from federal reserves that occur about 90 mi to the south in the northern Oregon Coast Range physiographic province. The federal reserves in the most northern part of the Oregon Coast Range are the second area of concern. It appears that regeneration timber harvesting on nonfederal land may be narrowing the dispersal connection to the rest of the Coast Range’s large federal reserved allocations. Finally, the southernmost large reserves, which are mainly located on the Mendocino National Forest in the California Klamath Mountains physiographic province, appear to occur in poor dispersal landscapes, and the Marin County northern spotted owl population, in particular, appears isolated at the extreme southern tip of the owl’s range (fig. 3-17).

Habitat Fragmentation At the range scale, core habitat accounted for about 19 and 29 percent of baseline nesting/roosting habitat within nonreserved and reserved allocations, respectively, indicating Figure 3-18—Nesting/roosting “core” habitat trends from 1994/96 to 2006/07 by physiographic province for reserved and nonre- that reserved allocations contain larger patches of suitable served federal lands. habitat. Between 1994/96 and 2006/07, the amount of core habitat on federal lands decreased by 6 percent at the The combination of core and core-edge habitat range scale, with 4.6 percent of this decrease occurring in constituted about 50 percent of the baseline nesting/ reserved allocations. The largest decrease (-20.6 percent) roosting habitat in nonreserved allocations and 61 percent occurred in the Oregon Klamath province and was largely in reserved allocations at the range scale, indicating that owing to the Biscuit Fire (fig. 3-18). The percentage loss of reserved allocations contain more contiguous habitat than core habitat by physiographic province shown in figure 3-18 nonreserved allocations (table 3-3). We report an average generally follows the same pattern among provinces as for rangewide decrease of 1 percent in these ratios, signifying nesting/roosting habitat loss (fig. 3-12); however, percentage a small but measureable increase in habitat fragmentation. of loss is larger for core habitat, because it is a subset of The largest decreases occurred within the federally reserved nesting/roosting habitat and confined to a smaller portion portions of the Klamath provinces in Oregon (-4.3 percent) of the landscape. and California (-2.7 percent) and California Cascades (-3.1 percent), again because of wildfires (table 3-3). 52 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Table 3-3–Habitat fragmentation status and trends based on the percentage of nesting/roosting habitat in core and core-edge habitat Reserved Nonreserved Physiographic province 1994/96 2006/07 Trend 1994/96 2006/07 Trend Percent Washington Olympic Peninsula 72.76 72.75 -0.01 33.22 32.86 -0.36 California Coast Range 68.77 68.81 0.04 46.64 47.19 0.55 Oregon Western Cascades 68.77 67.72 -1.05 59.51 58.39 -1.12 Oregon Coast Range 62.83 62.53 -0.30 39.41 37.77 -1.64 Oregon Klamath 61.69 57.35 -4.34 38.51 36.80 -1.71 Washington Western Cascades 59.23 59.06 -0.17 49.77 49.44 -0.33 Washington Eastern Cascades 55.69 54.39 -1.30 50.89 50.20 -0.69 Oregon Eastern Cascades 55.00 55.04 0.04 48.94 47.28 -1.66 California Cascades 51.70 50.67 -1.03 48.30 45.16 -3.14 California Klamath 49.01 46.31 -2.70 42.22 41.14 -1.08 Washington Western Lowlands 31.34 30.28 -1.06 0.00 0.00 0.00 Oregon Willamette Valley 25.74 25.93 0.19 27.83 26.13 -1.70

Note: Physiographic provinces are listed in order of least- to most-fragmented federal reserved land allocations based on the status in 1994/96. Negative trend values indicate increased fragmentation.

Discussion 10- to 12-year timeframe of our analysis data. The expecta- Substantial progress has been made in 5 years to overcome tion was that validation of habitat development would be some of the previous limitations of habitat monitoring part of the new habitat suitability maps developed by the (Davis and Lint 2005). Most importantly among these interagency monitoring program (Courtney et al. 2004). advancements is the development of a consistent set of However, validation of habitat development is a difficult vegetation data that now covers the entire range of the owl. task, and the transition of a forest age class or size class into As suspected by Davis and Lint (2005), the finer resolution the next higher class does not always equate to recruitment (in both spatial scale and attributes) of this new vegetation of owl habitat (Courtney et al. 2004). As seen from the data resulted in lower, but more accurate, estimates of the combinations of vegetation variables we used for habitat amount of northern spotted owl habitat in California. In modeling (app. A and table C-1 in app. C), the definition addition, the development of bookend maps (using the same of nesting/roosting habitat is not a simple combination of vegetation and modeling techniques) has increased our one or two attributes. In reality, it is much more complex, ability to detect trends of habitat losses and gains. For the and the transition of habitat from unsuitable to suitable first time, we can estimate not only habitat losses, but also likely happens over multiple decades (Courtney et al. 2004). habitat degradation—where habitat is altered by a distur- This was not the case for the younger forest types through bance, but still remains suitable for owl nesting and roost- which owls can disperse. We cautiously accounted for gains ing. The new LandTrendr change-detection data (Kennedy in dispersal habitats after examination of the dispersal et al. 2010) were critical for verifying the habitat losses we habitat maps on aerial imagery and through GIS analysis of detected with the bookends and also for assigning a cause changes in the tree diameter and canopy cover variables that for the habitat changes. were used in its definition. Although we were able to detect, measure, and report So, although Raphael et al. (1994a, 1994b) and Lint on nesting/roosting habitat loss and degradation, we were et al. (1999) did not expect to see any significant gains in not able to detect and measure its recruitment during the nesting/roosting habitat for a few decades, an examination of our habitat histograms (app. F) shows some gains in the

53 GENERAL TECHNICAL REPORT PNW-GTR-850

“marginal” suitability class, which is similar to dispersal Summary habitat (see table 3-1). Within the next three decades, the Rangewide owl habitat losses on federal lands were transition of habitat from the marginal suitability class to expected to be about 5 percent per decade, with a loss of the suitable habitat class may be detectable given current 2.5 percent from timber harvest (USDA and USDI 1994) remote sensing technology. In addition, the use of light and 2.5 percent from wildfire (FEMAT 1993). We report a detection and ranging (LIDAR) imagery, which is able to rangewide loss of 3.4 percent, between 1994/97 to 2006/07 map forest canopy biomass, height, and vertical distribution, and conclude that [rangewide] habitat is not declining may provide us the ability to detect and monitor changes faster than predicted under the Plan. Timber harvesting in the older stages of succession with improved accuracies accounted for 0.6 percent of this loss, insects and disease (Falkowski et al. 2009). 0.1 percent, and wildfire 2.7 percent of the habitat loss. Loss Maintaining and restoring habitats that keep owl popu- from timber harvesting is occurring at a fraction of what lations well connected across their range is a central goal of was predicted at Plan implementation, but habitat losses the Plan and should remain a priority. Our dispersal-capable from wildfire are very close to what was predicted (FEMAT landscape analysis was based on known linear dispersal 1993). Although rangewide habitat losses have not exceeded distances (Forsman et al. 2002, Lint et al. 2005), and the what was anticipated under the Plan, the trend of habitat analysis window we used to quantify amounts of dispersal loss has been greater than 5 percent per decade in some habitat across the landscape had a diameter of 31 mi. This physiographic provinces (i.e., Oregon Klamath). If localized distance exceeds both the mean natal dispersal distance for habitat losses continue at the current rates within some males and females (Forsman et al. 2002) and the root-mean- provinces, it is unclear what effect this may have on the square dispersal distance, which may be the more appropri- effectiveness of the Plan to maintain well-distributed and ate measure of gene flow (Barrowclough et al. 2005). Thus, connected populations of northern spotted owls throughout our results indicate that most of the large reserved network their entire range, specifically the assumption that the large is currently well connected (fig. 3-17) with a few exceptions, reserve network is resilient enough to incur these losses and such as the Olympic Peninsula, the northern Oregon Coast not result in isolation of population segments (Lint et al. Range, and the California Klamath, which we suggest might 1999). serve as focal areas for future studies on population con- Since implementation of the Plan, the majority of nectivity and genetics, particularly as recent genetic work habitat loss on federally administered lands has been caused suggests northern spotted owls have undergone population by wildfire, and most of that loss has occurred in reserved bottlenecks resulting in reduced genetic diversity in several allocations. This seems counter to the Plan’s goal of habitat parts of their range, including the northern Oregon Coast maintenance and restoration within the reserved network. Ranges, and the Klamath Mountains (Funk et al. 2010). However, the reserve network was designed to function Strong evidence for population bottlenecks in the Washing- despite losses to wildfire, which were anticipated (FEMAT ton eastern Cascades were also reported (Funk et al. 2010) 1993, Murphy and Noon 1992). Although Lint et al. (1999) consistent with recent population declines in that region assumed that habitat conditions within large reserves would (Anthony et al. 2006, Forsman et al. 2011), but there is no improve over time at a rate controlled by successional definitive evidence that dispersal habitat is limited (this processes in stands that are not currently nesting/roosting study) or that gene flow is restricted in that region habitat, they did not expect it to happen quickly, but over (Barrowclough et al. 2005). a period of several decades (Lint et al. 1999). Our latest

54 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

monitoring shows that maintenance of nesting/roosting References habitat within some of the large reserves is being challenged Akaike, H. 1974. A new look at the statistical model by the occurrence of large wildfires, and also that large- identification. IEEE Transactions on Automatic Control. scale restoration of reserved nesting/roosting habitat has not 19(6): 716–723. yet occurred. Anthony, R.G.; Forsman, E.D.; Franklin, A.B.; The monitoring assumption that habitat conditions Anderson, D.R.; Burnham, K.P.; White, G.C.; outside of reserved allocations would continue to decline Schwarz, C.J.; Nichols, J.D.; Hines, J.E.; Olson, because of timber harvesting and other habitat-altering G.S.; Ackers, S.H.; Andrews, L.W.; Biswell, B.L.; disturbances but would still facilitate owl movement across Carlson, P.C.; Diller, L.V.; Dugger, K.M.; Fehring, the landscape (Lint et al. 1999) is validated by the latest K.E.; Fleming, T.L.; Gearhardt, R.P.; Gremel, monitoring. The rate of nesting/roosting habitat loss outside S.A.; Gutiérrez, R.J.; Happe, P.J.; Herter, D.R.; of the reserves from timber harvesting has been lower Higley, J.M.; Horn, R.B.; Irwin, L.L.; Loschl, P.J.; than expected, and we observed both losses and gains in Reid, J.A.; Sovern, S.G. 2006. Status and trends in dispersal habitat. In our monitoring, we did not observe any demography of northern spotted owls, 1985–2003. isolation of owl population segments caused by large-scale Wildlife Monographs No. 163. Washington, DC: The disturbance; however, we did note both expansions and Wildlife Society. 48 p. contractions of dispersal-capable landscape and that some large reserves in portions of the range have poor dispersal Araújo, M.B.; Pearson, R.G.; Thuillers, W.; Erhard, M. conditions and might be focal areas for further investigation 2005. Validation of species-climate impact models under of population isolation studies. climate change. Global Change Biology. 11: 1504–1513. Although not included within the timeframe of this Baldwin, R.A. 2009. Use of maximum entropy modeling in latest monitoring analysis, the southern portion of the owl’s wildlife research. Entropy. 11: 854–866. range experienced another 615,000 ac (approx.) of wildfire Barrowclough, G.F.; Groth, J.G.; Mertz, L.A.; between 2008 and 2009, with most of it occurring within Gutiérrez, R.J. 2005. Genetic structure, introgression, reserved land use allocations. If this trend persists, the and a narrow hybrid zone between northern and actual decadal loss of habitat from wildfire will continue to California spotted owls (Strix occidentalis). Molecular push against the Plan’s assumption of 2.5 percent per decade Ecology. 14: 1109–1120. and, to reemphasize the point made at the beginning of this summary, may have unexpected consequences on the effec- Boyce, M.S.; Vernier, P.R.; Nielsen, S.E.; Schmiegelow, tiveness of portions of the large reserved network. Outside F.K.A. 2002. Evaluating resource selection functions. of the reserved network, the lack of timber harvesting in the Ecological Modelling. 157: 281–300. nonreserved allocations over the past 15 years has provided Braunisch, V.; Suchant, R. 2010. Predicting species some cushion from these losses. And finally, although we distributions based on incomplete survey data: the trade- still anticipate that recruitment of nesting/roosting habitat off between precision and scale. Ecography. 33: 826–840. from forest succession will eventually begin to offset habitat Brown, T.J.; Hall, B.L.; Mohrle, C.R.; Reinbold, losses from wildfire, forests grow slowly, and, where they H.J. 2002. Coarse assessment of federal wildland occur in landscapes prone to wildfire, the nesting/roosting fire occurrence data: report for the National Wildfire habitat conditions may take much longer to develop. Coordinating Group. Reno, NV: CEFA Report 02-04. Program for Climate, Ecosystem and Fire Applications, Desert Research Institute, Division of Atmospheric Sciences. 31 p.

55 GENERAL TECHNICAL REPORT PNW-GTR-850

Carey, A.B.; Horton, S.P.; Biswell, B.L. 1992. Northern Dugger, K.M.; Wagner, F.; Anthony, R.G.; Olson, G.S. spotted owls: influence of prey base and landscape 2005. The relationship between habitat characteristics character. Ecological Monographs. 62(2): 223–250. and demographic performance of northern spotted owls in southern Oregon. Condor. 107: 863–878. Carey, A.B.; Reid, J.A.; Horton, S.P. 1990. Spotted owl home range and habitat use in southern Oregon Coast Elith, J.; Graham, C.H.; Anderson, R.P.; Dudík, M.; Ranges. Journal of Wildlife Management. 54: 11–17. Ferrier, S.; Guisan, A.; Hijmans, R.J.; Huettmann, F.; Leathwick, J.R.; Lehmann, A.; Li, J.; Lohmann, Cohen, J. 1960. A coefficient of agreement for nominal L.G.; Loiselle, B.A.; Manion, G.; Moritz, C.; scales. Educational and Psychological Measurement. Nakamura, M.; Nakazawa, Y.; Overton, J.M.; 20(1): 37–46. Peterson, A.T.; Phillips, S.J.; Richardson, K.; Cohen, W.B.; Zhiqiang, Y.; Kennedy, R.E. 2010. Scachetti-Pereira, R.; Schapire, R.E.; Soberón, Detecting trends in forest disturbance and recovery J.; Williams, S.; Wisz, M.S.; Zimmermann, N.E. using yearly Landsat time series: 2. TimeSync—tools 2006. Novel methods improve prediction of species’ for calibration and validation. Remote Sensing of distributions from occurrence data. Ecography. Environment. 114(12): 2911–2924. 29: 129–151. Comer, P.; Faber-langendoen, D.; Evans, R.; Gawler, Elith, J.; Phillips, S.J.; Hastie, T.; Dudík, M.; Chee, S.; Josse, C.; Kittel, G.; Menard, S.; Pyne, M.; Y.E.; Yates, C.J. 2011. A statistical explanation of Reid, M.; Schulz, K.; Snow, K.; Teague, J. 2003. MaxEnt for ecologists. Diversity and Distributions. Ecological systems of the United States: a working 17: 43–57. classification of U.S. terrestrial systems. Arlington, VA: Engler, R.; Guisan, A.; Rechsteiner, L. 2004. An NatureServe. 75 p. http://www.natureserve.org/library/ improved approach for predicting the distribution of usEcologicalsystems.pdf. (18 December 2010). rare and endangered species from occurrence and Coops, N.C.; Gillandersa, S.N.; Wulderb, M.A.; Gergelc, pseudoabsence data. Journal of Applied Ecology. S.E.; Nelsond, T.; Goodwin, N.R. 2010. Assessing 41: 263–274. changes in forest fragmentation following infestation Falkowski, M.J.; Evans, J.S.; Martinuzzi, S.; Gessler, using time series Landsat imagery. Forest Ecology and P.E.; Hudak, A.T. 2009. Characterizing forest succes- Management. 259: 2355–2365. sion with LIDAR data: an evaluation for the inland Courtney, S.P.; Blakesley, J.A.; Bigley, R.E.; Cody, Northwest, USA. Remote Sensing of Environment. M.L.; Dumbacher, J.P.; Fleischer, R.C.; Franklin, 113: 946–956. A.B.; Franklin, J.F.; Gutiérrez, R.J.; Marzluff, J.M.; Fielding, A.H.; Bell, J.F. 1997. A review of methods for Sztukowski, L. 2004. Scientific evaluation of the status assessment of prediction errors in conservation presence/ of the northern spotted owl. Portland, OR: Sustainable absence models. Environmental Conservation. 24: Ecosystem Institute. 348 p. + appendixes. 38–49. Davis, R.; Lint, J. 2005. Habitat status and trend. In: Forest Ecosystem Management Assessment Team Lint, J., tech. coord. Northwest Forest Plan—the first [FEMAT]. 1993. Forest ecosystem management: an 10 years (1994–2003): status and trends of northern ecological, economic, and social assessment. Portland, spotted owl populations and habitat. Gen. Tech. Rep. OR: U.S. Department of Agriculture; U.S. Department PNW-GTR-648. Portland, OR: U.S. Department of of the Interior [and others]. [Irregular pagination]. Agriculture, Forest Service, Pacific Northwest Research Station: 21–82.

56 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Forsman, E.D.; Anthony, R.G.; Dugger, K.M.; Glenn, Gauch, H.G. 1982. Multivariate analysis in community E.M.; Franklin, A.B.; White, G.C.; Schwarz, C.J.; ecology. Cambridge, England: Cambridge University Burnham, K.P.; Anderson, D.R.; Nichols, J.D.; Hines, Press. 312 p. J.E.; Lint, J.B.; Davis, R.J.; Ackers, S.H.; Andrews, Gillison, A.N.; Brewer, K.R.W. 1985. The use of gradient L.S.; Biswell, B.L.; Carlson, P.C.; Diller, L.V.; directed transects or gradsects in natural resource Gremel, S.A.; Herter, D.R.; Higley, J.M.; Horn, R.B.; surveys. Journal of Environmental Management. 20: Reid, J.A.; Rockweit, J.; Schaberl, J.; Snetsinger, T.J.; 103–127. Sovern, S.G. 2011. Population demography of northern spotted owls. Studies in Avian Biology 40. Berkeley, CA: Glenn, E.M.; Hansen, M.C.; Anthony, R.G. 2004. University of California Press. 106 p. Spotted owl home range and habitat use in young forests of western Oregon. Journal of Wildlife Management. 68: Forsman, E.D.; Anthony, R.G.; Reid, J.A.; Loschl, P.J.; 33–50. Sovern, S.G.; Taylor, M.; Biswell, B.L.; Ellingson, A.; Meslow, E.C.; Miller, G.S.; Swindle, K.A.; Thrailkill, Guisan, A.; Zimmermann, N.E. 2000. Predictive habitat J.A.; Wagner, F.F.; Seaman, D.E. 2002. Natal and distribution models in ecology. Ecological Modelling. breeding dispersal of northern spotted owls. Wildlife 135: 147–186. Monographs No. 149. Washington, DC: The Wildlife Gutíerrez, R.J.; Solis, D.M.; Sisco, C. 1984. Habitat Society. 35 p. ecology of the spotted owl in northwestern California: Forsman, E.D.; Meslow, E.C.; Wight, H.M. 1984. implications for management. In: Proceedings of the Distribution and biology of the spotted owl in Oregon. 1983 convention of the Society of American Foresters. Wildlife Monographs. 87: 1–64. New forests for a changing world. Bethesda, MD: Society of American Foresters: 368–373. Franklin, A.B.; Anderson, D.R.; Gutiérrez, R.J.; Burnham, K.P. 2000. Climate, habitat quality, Hais, M.; Jonášová, M.; Langhammer, J.; Kučera, T. and fitness in northern spotted owl populations in 2009. Comparison of two types of forest disturbance northwestern California. Ecological Monographs. 70(4): using multitemporal Landsat TM/ETM + imagery and 539–590. field vegetation data. Remote Sensing of Environment. 113: 835–845. Franklin, J.F.; Dyrness, C.T. 1973. Natural vegetation of Oregon and Washington. Gen. Tech. Rep. PNW-8. Hamer, T.E.; Seim, S.G.; Dixon, K.R. 1989. Northern Portland, OR: U.S. Department of Agriculture, Forest spotted owl and northern barred owl habitat use and Service, Pacific Northwest Forest and Range Experiment home range size in northwest Washington: preliminary Station. 417 p. report. 64 p. Unpublished report. On file with: Washington Department of Fish and Wildlife, 1111 Frolking, S.; Palace, M.W.; Clark, D.B.; Chambers, Washington Street Southeast, Olympia, WA 98501. J.Q.; Shugart, H.H.; Hurtt, G.C. 2009. Forest disturbance and recovery: a general review in the context Herold, N.D.; Koeln, G.; Cunnigham, D. 2003. of spaceborne remote sensing of impacts on aboveground Mapping impervious surfaces and forest canopy biomass and canopy structure. Journal of Geophysical using classification and regression tree (CART) Research. 114. 27 p. analysis. In: Proceedings of the American Society for Photogrammetry and Remote Sensing 2003 annual Funk, W.C.; Forsman, E.D.; Johnson, M.; Mullins, conference, Anchorage, Alaska. 7 p. http://nemo.uconn. T.D.; Haig, S.M. 2010. Evidence for recent population edu/tools/impervious_surfaces/pdfs/Herold_etal_2003. bottlenecks in northern spotted owls (Strix occidentalis pdf. (5 January 2011). caurina). Conservation Genetics. 11: 1013–1021.

57 GENERAL TECHNICAL REPORT PNW-GTR-850

Hirzel, A.H.; Hausser, J.; Chessel, D.; Perrin, N. 2002. Kennedy, R.E.; Yang, Z.; Cohen, W.B. 2010. Detecting Ecological-niche factor analysis: how to compute trends in forest disturbance and recovery using yearly habitat-suitability maps without absence data. Ecology. Landsat time series: 1. LandTrendr—temporal segmenta- 83: 2027–2036. tion algorithms. Remote Sensing of Environment. Hirzel, A.H.; Le Lay, G. 2008. Habitat suitability 114(12): 2897–2910. modelling and niche theory. Journal of Applied Ecology. Lint, J.; Mowdy, J.; Ried, J.; Forsman, E.; Anthony, R. 45: 1372–1381. 2005. Owl movement. In: Lint, J., tech. coord. Northwest Hirzel, A.H.; Le Lay, G.; Helfer, V.; Randin, C.; Guisan, Forest Plan—the first 10 years (1994–2003): status and A. 2006. Evaluating the ability of habitat suitability trends of northern spotted owl populations and habitat. models to predict species presences. Ecological Gen. Tech. Rep. PNW-GTR-648. Portland, OR: U.S. Modelling. 199: 142–152. Department of Agriculture, Forest Service, Pacific Northwest Research Station: 83–87. Huff, M.; Raphael, M.G.; Miller, S.L.; Nelson, S.K.; Baldwin, J., tech. coords. 2005. Northwest Forest Lint, J.; Noon, B.; Anthony, R.; Forsman, E.; Raphael, Plan—the first 10 years (1994–2003): status and trends M.; Collopy, M.; Starkey, E. 1999. Northern spotted of populations and nesting habitat for the marbled owl effectiveness monitoring plan for the Northwest murrelet. Gen. Tech. Rep. PNW-GTR-650. Portland, OR: Forest Plan. Gen. Tech. Rep. PNW-GTR-440. Portland, U.S. Department of Agriculture, Forest Service, Pacific OR: U.S. Department of Agriculture, Forest Service, Northwest Research Station. 149 p. Pacific Northwest Research Station. 43 p. Hutchinson, G.E. 1957. Concluding remarks. Cold Spring Lint, J.B., tech. coord. 2005. Northwest Forest Plan—the Harbor Symposium. Quantitative Biology. 22: 415–427. first 10 years (1994–2003): status and trends of northern spotted owl populations and habitat. Gen. Tech. Rep. Jenness, J. 2003. Mahalanobis distances (mahalanobis.avx) PNW-GTR-648. Portland, OR: U.S. Department of extension for ArcView 3.x, Jenness Enterprises. http:// Agriculture, Forest Service, Pacific Northwest Research www.jennessent.com/arcview/mahalanobis.htm. Station. 176 p. (18 December 2010). Liu, C.; Berry, P.M.; Dawson, T.P.; Pearson, R.G. 2005. Jiménez-Valverde, A.; Nakazawa, Y.; Lira-Noriega, Selecting thresholds of occurrence in the prediction of A.; Peterson, A.T. 2009. Environmental correlation species distributions. Ecography. 28: 385–393. structure and ecological niche model projections. Biodiversity Informatics. 6: 28–35. Manley, B.; McDonald, L.; Thomas, D. 1993. Resource selection by animals: statistical design and analysis for Kennedy, R.E. 2010. Personal communication. Assistant field studies. London: Chapman and Hall. 177 p. professor, senior researcher and director, Laboratory for Applications of Remote Sensing in Ecology, Oregon McArdle, R.E.; Meyer, W.H.; Bruce, D. 1961. The yield State University, College of Forestry, Forest Ecosystems of Douglas-fir in the Pacific Northwest. Rev. Tech. Bull. and Society. Corvallis, OR 97331. 201. Washington, DC: U.S. Department of Agriculture, Forest Service. 74 p. Kennedy, R.E.; Cohen, W.B.; Schroeder, T.A. 2007. Trajectory-based change-detection for automated Miller, G.S. 1989. Dispersal of juvenile northern spotted characterization of forest disturbance dynamics. Remote owls in western Oregon. Corvallis, OR: Oregon State Sensing of Environment. 110: 370–386. University. 139 p. M.S. thesis.

58 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Miller, G.S.; Small, R.J.; Meslow, E.C. 1997. Habitat Ohmann, J.L.; Gregory, M.J.; Roberts, H.M. 2010. selection by spotted owls during natal dispersal in GNN “bookend” maps of forest vegetation for NWFP western Oregon. Journal of Wildlife Management. 61: effectiveness monitoring. 4 p. Unpublished report. 140–150. On file with: U.S. Department of Agriculture, Forest Moeur, M.; Ohmann, J.L.; Kennedy, R.E.; Cohen, W.B.; Service, 333 SW First Ave., Portland, OR 97208-3623. Gregory, M.J.; Yang, Z.; Roberts, H.M.; Spies, T.A.; Ohmann, J.L.; Gregory, M.J.; Spies, T.A. 2007. Influence Fiorella, M. [In press]. Northwest forest plan—the of environment, disturbance, and ownership on forest first 15 years (1994–2008): status and trends of late- vegetation of coastal Oregon. Ecological Applications. successional and old-growth forests. Gen. Tech. Rep. 17: 18–33. Portland, OR: U.S. Department of Agriculture, Forest Olson, G.S.; Glenn, E.M.; Anthony, R.G.; Forsman, Service, Pacific Northwest Research Station. E.D.; Reid, J.A.; Loschl, P.J.; Ripple, W.J. 2004. Moeur, M.; Spies, T.A.; Hemstrom, M.; Alegria, J.; Modeling demographic performance of northern spotted Browning, J.; Cissel, J.; Cohen, W.B.; Demeo, T.E.; owls relative to forest habitat in Oregon. Journal of Healey, S.; Warbington, R. 2005. Northwest Forest Wildlife Management. 68: 1039–1053. Plan—the first 10 years (1994–2003): status and trend Omernik, J.M. 1987. Ecoregions of the conterminous of late-successional and old-growth forest. Gen. Tech. United States (map supplement) [1:750,000]. Annals Rep. PNW-GTR-646. Portland, OR: U.S. Department of of the Association of American Geographers. 77(1): Agriculture, Forest Service, Pacific Northwest Research 118–125. Station. 142 p. Peterson, A.T.; Papeş, M.; Eaton, M. 2007. Transferability Moeur, M.; Stage, A.R. 1995. Most similar neighbor: and model evaluation in ecological niche modeling: an improved sampling inference procedure for natural a comparison of GARP and MaxEnt. Ecography. 30: resource planning. Forest Science. 41: 337–359. 550–560. Morrison, M.L.; Marcot, B.G.; Mannan, R.W. 1992. Phillips, S.J. 2008. Transferability, sample selections Wildlife-habitat relationships: concepts and applications. bias and background data in presence-only modeling: Madison, WI: The University of Wisconsin Press. 364 p. a response to Peterson et al. (2007). Ecography. 31: Murphy, D.D.; Noon, B.R. 1992. Integrating scientific 272–278. methods with habitat conservation planning: reserve Phillips, S.J. 2010. Personal communication. Co-developer design for northern spotted owls. Ecological Applica- of MaxEnt. AT&T Labs–Research, 180 Park Avenue, tions. 2(1): 3–17. Florham Park, NJ 07932. Ohmann, J.L.; Gregory, M.J. 2002. Predictive mapping Phillips, S.; Dudík, M. 2008. Modeling of species of forest composition and structure with direct gradient distributions with MaxEnt: new extensions and a analysis and nearest-neighbor imputation in coastal comprehensive evaluation. Ecography. 31: 161–175. Oregon, U.S.A. Canadian Journal of Forest Research. 32: 725–741. Phillips, S.J.; Anderson, R.P.; Shapire, R.E. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling. 190: 231–259.

59 GENERAL TECHNICAL REPORT PNW-GTR-850

Phillips, S.J.; Dudík, M.; Elith, J.; Graham, C.H.; Raphael, M.G.; Young, J.A.; McKelvey, K.; Galleher, Lehmann, A.; Leathwick, J.; Ferrier, S. 2009. Sample B.M.; Peeler, K.C. 1994b. Addendum to “A simulation selection bias and presence-only distribution models: analysis of population dynamics of the northern spotted implications for background and pseudo-absence data. owl in relation to forest management objectives. Ecological Applications 19: 181–197. Appendix J3. In: U.S. Department of Agriculture, Forest Pierce, K.B.J.; Ohmann, J.L.; Wimberly, M.C.; Service; U.S. Department of the Interior, Bureau of Land Gregory, M.J.; Fried, J.S. 2009. Mapping wildland Management. Final supplemental environmental impact fuels and forest structure for land management: a statement on management of late-successional and old- comparison of nearest-neighbor imputation and other growth forests within the range of the northern spotted methods. Canadian Journal of Forest Research. 39: owl. 16 p. 1901–1916. Soille, P.; Vogt, P. 2009. Morphological segmentation Randin, C.F.; Dirnbo, T.; Dullinger, S.; Zimmermann, of binary patterns. Pattern Recognition Letters. 30: N.E.; Zappa, M.; Guisan, A. 2006. Are niche-based 456–459. species distribution models transferable in space? Journal Stauffer, D.F. 2002. Linking populations and habitat: of Biogeography. 33: 1689–1703. Where have we been? Where are we going? In: Scott, Raphael, M. 2006. Conservation of listed species: the J.M.; Heglund, P.; Morrison, M., eds. Predicting species northern spotted owl and marbled murrelet. In: Haynes, occurrences: issues of accuracy and scale. Washington, R.; Bormann, B.; Lee, D.; Martin, J., eds. Northwest DC: Island Press: 53–61. Forest Plan—the first 10 years (1994–2004): synthesis of Staus, N.; Strittholt, J.; Dellasala, D.; Robinson, R. 2002. monitoring and research results. Gen. Tech. Rep. PNW- Rate and patterns of forest disturbance in the Klamath- GTR-651. Portland, OR: U.S. Department of Agriculture, Siskiyou ecoregion, USA between 1972 and 1992. Forest Service, Pacific Northwest Research Station: Landscape Ecology. 17: 455–470. 117–144. Swets, J. 1988. Measuring the accuracy of diagnostic Raphael, M.G.; Young, J.A.; McKelvey, K.; Galleher, systems. Science. 240: 1285–1293. B.M.; Peeler, K.C. 1994a. A simulation analysis of ter Braak, C.J.F. 1986. Canonical correspondence population dynamics of the northern spotted owl in analysis: a new eigenvector technique for multivariate relation to forest management objectives. In: Final direct gradient analysis. Ecology. 67: 1167–1179. supplemental environmental impact statement on management of late-successional and old-growth Thomas, J.W.; Forsman, E.D.; Lint, J.B.; Meslow, E.C.; forests within the range of the northern spotted owl. Noon, B.R.; Verner, J. 1990. A conservation strategy Portland, OR: U.S. Department of Agriculture, Forest for the northern spotted owl: a report of the Interagency Service; U.S. Department of the Interior, Bureau of Land Scientific Committee to address the conservation of the Management. Vol. II, app. J-3 northern spotted owl. Portland, OR: U.S. Department of Agriculture, Forest Service; U.S. Department of the Interior, Bureau of Land Management, Fish and Wildlife Service, National Park Service. 427 p.

60 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Turner, M.G.; Dale, V.H.; Gardner, R.H. 1989. Predicting Vogt, P.; Riitters, K.H.; Estreguil, C.; Kozak, J.; Wade, across scales: theory development and testing. Landscape T.G.; Wickham, J.D. 2007. Mapping spatial patterns Ecology. 3: 245–252. with morphological image processing. Landscape U.S. Department of Agriculture, Forest Service; Ecology. 22: 171–177. U.S. Department of the Interior, Bureau of Land Wiley, E.O.; McNyset, K.M.; Peterson, A.T.; Robins, Management [USDA and USDI]. 1994. Final C.R.; Stewart, A.M. 2003. Niche modeling and supplemental environmental impact statement on geographic range predictions in the marine environment management of habitat for late-successional and old- using a machine learning algorithm. Oceanography. 16: growth forest related species within the range of the 120–127. northern spotted owl. Volumes I-II + Record of Decision. Williams, P.H.; Margules, C.R.; Hilbert, D.W. 2002. Vogelmann, J.E.; Howard, S.M.; Yang, L.; Larson, C.R.; Data requirements and data sources for biodiversity Wylie, B.K.; Van Driel, J.N. 2001. Completion of the priority area selection. Journal of Biosciences. (Suppl. 2). 1990’s national land cover data set for the conterminous 27(4): 327–338. United States. Photogrammetric Engineering and Zabel, C.J.; Dunk, J.R.; Stauffer, H.B.; Roberts, L.M.; Remote Sensing. 67: 650–662. Mulder, B.S.; Wright, A. 2003. Northern spotted owl habitat models for research and management application in California (USA). Ecological Applications. 13(4): 1027–1040. Kristian Skybak

61 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Chapter 4: Large Wildfires Within the Owl’s Range Raymond J. Davis, William C. Aney, Louisa Evers, and Katie M. Dugger

Introduction 1992) used this and other information to further subdivide When Franklin and Dyrness created their map of physio- the range into 12 physiographic provinces, which currently graphic provinces in 1973, they noted that the lines drawn provide the framework for monitoring the Northwest Forest to reduce the complexity of large geographic areas into Plan (the Plan) (FEMAT 1993, Lint et al. 1999). More recent more manageable proportions are sometimes arbitrary, attempts to map the “dry, fire-prone” portion of the owl’s whereas in nature the transition from one condition to range (Healey et al. 2008, Rapp 2005, Spies et al. 2006) another is often gradual. A modified version of the Franklin are mainly delineated along these physiographic province and Dyrness (1973) physiographic provinces was used to di- boundary lines, which were not drawn specifically to define vide the northern spotted owl’s (Strix occidentalis caurina) the underlying nature of wildfire within the owl’s range. range, which covers 57 million ac that stretch from Canada The result is a line that often shifts, sometimes consider- to northern California, into 10 areas that represented dif- ably, between mapping efforts (fig. 4-1). ferent forest vegetation and environmental characteristics This desire to map fire-prone areas in the owl’s range (Thomas et al. 1990). Agee and Edmonds (1992) made the stems from a concern by many that wildfire will destroy first attempt to delineate fire disturbance regimes within the spotted owl habitat. The recent increase in frequency owl’s range during the initial stages of northern spotted owl of large wildfire occurrence (and area burned) since the recovery planning. The spotted owl recovery team (USDI mid-1980s in the Western United States (Schwind 2008, Westerling et al. 2006), and within the owl’s range (fig. 4-2)

Figure 4-1—Various depictions of the “fire-prone” areas within the range of the northern spotted owl (Agee and Edmonds 1992, Rapp 2005, Spies et al. 2006).

63 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-2—Frequency histogram of acres burned by wildfires within the range of the northern spotted owl between 1970 and 2009 (data sources from large wildfire data from this analysis). has only heightened this concern. There is also evidence accessible in these high-severity burn patches (Bond et al. that along with this increased frequency there has also been 2009, Clark 2007, Franklin et al. 2000). Thus, although an increase in the amount of high-severity wildfire (Miller stand-replacing wildfires certainly remove nesting/roost- et al. 2009, Schwind 2008; but see Hanson et al. 2009). ing habitats described in chapter 3, they may not prevent However, evidence from recent studies reveals that the foraging by owls, and only a very large fire that creates a effects of wildfire on owl habitat and demography are mixed large-scale loss of forest canopy and habitat would have a (Bond et al. 2009, Clark 2007). In the short term, large significant effect on owl demography and dispersal (see the wildfires may be detrimental to spotted owls by decreasing discussion on dispersal habitat in chapter 3). Much more survival and occupancy rates because high-severity1 fire research is needed to fully understand the effects of wildfire that caused loss and fragmentation of suitable nesting and frequency and severity on owls and their prey sources (see roosting habitat contributed to existing spotted owl sites chapter 5 in this report), but some adaptation to wildfire is becoming unoccupied (Clark 2007). In addition, California expected given that this species has evolved with it in some spotted owls avoided roosting [breeding season] in forests parts of its range. that had experienced moderate to high-severity2 fire effects Although the relationship between wildfire frequency and nested only in stands that were unburned or had expe- and severity on owl demography is not fully understood, rienced low- to moderate-severity fire (Bond et al. 2009). habitat loss is the primary reason for the owl’s decline and However, spotted owls did forage in areas of high-severity subsequent listing as “threatened” under the Endangered fire, possibly because prey species are more abundant and Species Act (USDI 1990). The habitat monitoring results presented in chapter 3 (this report) identified wildfire as the 1 Clark (2007) defined high-severity as > 70 percent of the overstory removed by fire. leading cause of current spotted owl nesting and roosting 2 Bond et al. (2009) defined high-severity as areas where dominant habitat loss (3.4 percent) and its future recruitment on vegetation had high to complete mortality owing to fire. federal lands. This was also the finding in the 10-year

64 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

monitoring report (Davis and Lint 2005), and since in the multidimensional environmental space (the number completion of that report, several more large wildfires have of dimensions are based on the number of environmental occurred within the owl’s range and more nesting/roost- variables used to describe the niche) that permits positive ing habitat has been lost. Thus, loss of habitat to wildfire growth (Hutchinson 1957). Habitat suitability models are remains a significant concern for the management and operational applications of the ecological niche, and use conservation of the spotted owl. In response, the current multiple environmental variables to predict the likelihood of species recovery planning process for the owl (USDI 2008) species occurrence (Hirzel and Le Lay 2008). established working groups to develop recovery actions for Based on our understanding of northern spotted owl fire-prone areas based on the current map of physiographic ecology, we expect them to nest in landscapes that are provinces (USDI 1992). heavily forested with older or structurally diverse stands Here we present a novel modeling method to map areas of conifer with relatively closed canopies (see chapter 3 within the owl’s range that are prone to large wildfires. in this report). We call this combination of environmental The result is a rangewide map of likelihood (or suitability) conditions owl “habitat.” Similarly, environmental condi- gradients for large wildfire occurrence. Instead of using tions commonly associated with large wildfires include physiographic province boundaries to define fire-prone steep slopes, warm and dry aspects, hot and dry weather, areas within the owl’s range, the gradient map is further and limited access for ground-based firefighting resources classified into a binary map that we believe better represents (hand crews, engines, etc.). These have long been identified the fire-prone areas. However, the raw model output (fig. in the literature as key elements in the development of large 4-3) maintains the gradual transitions from one condition wildfires (Albini 1976, Albini et al. 1982, Brown and Davis to another so succinctly alluded to by Franklin and Dyrness 1973, Countryman 1964, Deeming et al. 1977, Garfin and (1973). Morehouse 2001, Gisborne 1936, Hayes 1941, Rothermel 1983, Schroeder and Buck 1970, Scott and Reinhardt 2001, Methods and Data Sources Sugihara et al. 2006, Van Wagner 1977); in decision- There are several modeling approaches and methods avail- support planning tools for wildfire response such as the able for modeling spatial distributions of environmental National Fire Management Analysis System (NFMAS) and phenomenon, each with their own strengths and weaknesses its successor, Fire Program Analysis (FPA); and in practice. (Guissan and Zimmermann 2000). A recent paper by Elith It is no surprise that wildfires grow rapidly and become et al. (2011) summarizes many of these issues, including larger in landscapes that have an abundance of these condi- an ecological explanation of MaxEnt (Phillips and Dudík tions. The combination of these environmental conditions 2008, Phillips et al. 2006) and discussion on the issue of might also be considered a “habitat,” not for an animal, but using presence-only versus presence-absence data (also see one that is suitable for large wildfires as alluded to by Pyne page 35 of Davis and Lint 2005). For consistency, we chose (2001, 2004). The analogy of wildfire as a “living organ- to use MaxEnt, the same modeling tool used for mapping ism” is not unheard of (Bond and Keeley 2005, Parisien and spotted owl habitat suitability in chapter 3 (this report), to Moritz 2009), and it seems reasonable that the principles for model and map wildfire suitability (fig. 4-3). This spatial describing the niche of a plant or animal species should be distribution modeling software is commonly used to create no different than for defining the “niche” of large wildfires, predictive maps of habitat suitability (or likelihood of use) or for that matter any other natural phenomenon that is based on species location data and a set of environmental associated with unique combinations of environmental predictor variables that contribute to the definition of the conditions. species’ niche (Phillips et al. 2006). The term “niche” is Our ability to accurately map the environmental used to describe the environmental requirements needed for conditions that constitute the niche allows us to use model- a species to exist (Grinnell 1917). It is the “hypervolume” ing software to map the pattern of the relationship between

65 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-3—Although large (≥ 1,000 ac) wildfires are known throughout the entire range of the owl, this wildfire suitability map represents the likelihood for occurrence of these fires based on three decades of large wildfire occurrence and the underlying combination of “fire environment” variables from where they occurred (see fig. 4-4).

66 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

these environmental factors and large wildfire occurrence. background consisted only of forest-capable areas, which This approach has been used recently to model wildfire’s represent “vegetation” in the larger spatial scale. Because broad geographical distribution patterns across the con- forest fires are what we were attempting to model, the use terminous United States, the state of California, and five of this modeling background allowed us to confine the wildfire-prone ecoregions within California (Parisien and interactions of environmental variables to locations where Moritz 2009). To our knowledge, this marks the first time forest vegetation and fuels occur. An advantage of not using that “habitat suitability” software was used to map spatial a fuel variable is that we avoided the difficulties that arise patterns of wildfire likelihood over large landscapes as a in accurately mapping them (Stratton 2006). Fuels are a function of multiple environmental variables. Coarse-scale dynamic component of the ecosystem, very temporal in maps of global fire patterns that discriminated between nature and always changing in response to forest succession “fire-prone” and “fire-free” areas of the world were also and disturbances (Agee 1993). The inclusion of a fuel vari- produced using similar methods (Krawchuk et al. 2009). able would produce a map that would only be good as long Maps produced by this method have been called “wildfire as the fuel condition remained exactly as modeled. Instead, suitability” maps (Parisien and Moritz 2009), and this is the we wanted to produce a model that was relatively stable, term we use to describe our map (fig. 4-3). and based on the underlying conditions of topography and It is not uncommon for wildfires that range from 500 to climate that support large wildfires. 1,000 ac and greater to be defined as “large” in the recent The set of environmental variables we used for model- fire ecology literature (Potter 1996, Eidenshink et al. 2007, ing was based on fire climate4 and environment relation- Preisler and Westerling 2007, Westerling et al. 2003). In the ships in the literature and on expert advice (fig. 4-4, app. G). 10-year report, a “large wildfire” was defined as a fire that Matching the temporal scale of these environmental data would affect multiple owl territories (Davis and Lint 2005). with the fire training data was an important factor. Fire Here we define “large wildfire” as one that exceeds 1,000 climate variables were derived from “parameter elevated ac, which is larger than the estimated size of a northern regression on independent slope model” (PRISM) maps spotted owl home range core area3 throughout most of (Oregon Climate Service 2008) that provide averaged its range (Bingham and Noon 1997, Courtney et al. 2004, weather conditions between 1971 and 2000. This timeframe USDI and USDA 2008). coincides with the 1970 to 2002 timeframe of the fire training data set. As our fire climate variables, we initially chose Environmental Data average maximum temperature in August and summer At an intermediate spatial scale, weather and topography moisture stress (the ratio of summer temperature and make up two legs of the fire behavior (or environment) precipitation). However, because of the high correlation triangle (Agee 1993, Countryman 1966), whereas at the between these two variables (r > 0.7), we replaced the larger (regional) spatial scale, climate, ignitions, and broad moisture stress variable with a summer precipitation vari- vegetation patterns define fire regimes (see fig. 1 in Parisien able, which is the average amount of precipitation that fell and Mortiz 2009). Our spatial scale of modeling combines between May and September, corresponding to the average both the intermediate and regional scales, and our set of fire season. environmental data reflects this, with the exception of fuels Lightning is the primary ignition source for wildfires and vegetation variable groups. We did not include any around the world (Agee 1993) including the forested regions fuel variables in our modeling, but the model’s geographic of the Pacific Northwest, especially when it occurs without

3 An area of concentrated use within a home range that commonly 4 Defined by the National Wildfire Coordinating Group (NWCG) includes nest sites, roost sites, refuges, and regions with the most as a “composite pattern of weather elements over time that affect dependable food sources (Kaufmann 1962, Samuel et al. 1985). fire behavior in a given region.”

67 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-4—Environmental variables used in the model to define the “niche” of large wildfires in the owl’s range.

68 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

significant rainfall (Rorig and Ferguson 1999). Based on responsible for the vast majority of area burned each year. data from the Climate, Ecosystem, and Fire Applications For example, of the roughly 25,000 lightning-ignited (CEFA) Program (Brown et al. 2002), lightning was the wildfires recorded within the owl’s range between 1970 to cause for approximately 25,000 wildfires within the range 2002, less than 1 percent were ≥1,000 ac; but these fires ac- of the owl from 1970 to 2002. Lightning accounted for 68 counted for 96 percent of the total 2.5 million ac that burned percent of the wildfires that grew to larger than 1,000 ac and (based on CEFA data) (Brown et al. 2002). This pattern accounted for 75 percent of the total area burned within the of large areas of land being burned by a small percentage owl’s range from 1994 through 2002 (Davis and Lint 2005). of large wildfires is a global phenomenon that fits power The geographic patterns of lightning-ignited wildfires in law distributions (Cui and Perera 2008, Stocks et al. 2003, the Pacific Northwest are similar today to what they were Westerling and Bryant 2008). It therefore made more sense throughout the 1900s (Agee 1993, Komarek 1967, Morris to focus our modeling on large wildfires because of their 1934, Rorig and Ferguson 1999, Sensenig 2002). Therefore, disproportionate effect on the environment. a lightning-ignited fire density map was created using the To train the distributional model, the spatial point CEFA data from 1970 to 2002 and included as one of the locations where large wildfires have occurred are linked environmental variables. to the underlying combinations of environmental variable Topographic variables for elevation, slope, and aspect grid cells over which they lay. This relationship between were also used in the model. Elevation provides an environ- fire occurrence and environmental gradients is then extra- mental gradient that relates to local climate conditions and polated to the rest of the modeled region to “score” envi- vegetation zones, which can affect fire behavior and growth ronmental conditions based on their similarity to where (Hayes 1941, Rothermel 1983). Slope is related to fire spread the training data occur. To create a point layer representing rate, and its orientation, or aspect, relates to the amount of large wildfires, we assembled 250 polygons of large wildfire solar radiation, which also affects the local microclimate perimeters that, in total, burned about 2.6 million ac of and vegetation. Southerly aspects in the Northern Hemi- forest lands across the owl’s range between 1970 and 2002. sphere usually receive more annual solar radiation and Using a geographic information system (GIS), we overlaid are hotter and drier than northerly aspects. We used the these polygons on a grid of randomly generated points that potential relative radiation (PRR) index developed by Pierce was produced using Hawth’s Tools (Beyer 2009). Each grid et al. (2005) as a more realistic measure for solar radiation point was separated by 1.6 mi to reduce spatial autocorrela- than simple aspect. tion issues, as the modeling environmental variables were The spatial resolution of our environmental data was averaged over a 0.7-mi radius that covered about 1,000 ac, 250- by 250-m (15-ac) pixels, which was averaged within a representing a “large wildfire unit” (fig. 4-5). A total of 1,000-ac circular moving window to correspond with our 1,499 random grid points occurred within a large-wildfire minimum definition of a large wildfire. All “nonforested” perimeter; of these, 104 (about 7 percent) were within (i.e., water, rock, etc.) areas were “masked out” to constrain overlapping wildfire perimeters, representing sites that had the modeling background to only those areas where large been burned twice during the 32 years represented by our wildfires are possible. All variables were analyzed for spa- training data. Because these points represent separate large tial correlations and one variable was dropped or replaced wildfire occurrence from different years, they were infor Pearson correlations > 0.7. cluded as additional points in the training data set for a total of 1,603 training points. We also generated an independent Large Wildfire Data model-testing data set in the same manner, using 146 large We chose to train our model using historical occurrence wildfires that had burned 1.4 million ac between 2003 and data from only large wildfires (as defined above). Wild- 2009 (n = 903). fires of this size are relatively rare occurrences, but are

69 GENERAL TECHNICAL REPORT PNW-GTR-850

avoiding features that might allow overfitting the model to the data (i.e., threshold). Phillips and Dudík (2008) defined the logistic output of their modeling software as an “estimate” of probability of presence, conditioned on the environmental variables used in the modeling. In our case, we used only training data from large wildfires and environmental variables that are commonly associated with wildfire ignition and growth (Albini 1976, Deeming et al. 1977, Gisborne 1936, Hayes 1941, Rothermel 1991). Therefore, the model’s logistic output represents a scale of probability (from low to high) of a large wildfire occurring as a function of physical, topographic, climatic, and fire ignition history patterns in the owl’s range. Where combinations of these variables are more similar to where large wildfires have occurred in the Figure 4-5—Model training and testing data came from grid past, the logistic probability values will be higher. Likewise, points that occurred within large-fire perimeters. Points were spaced by 2.5 km to reduce spatial autocorrelation of environmen- underlying patterns of environmental variables that do not tal data, which were averaged within 1,000-ac circles as shown in commonly occur where large wildfires have burned will this figure. have lower probability values. It is likely that without wildfire suppression, there We ran 10 bootstrapped model replicates using half of would have been more large wildfires (Sensenig 2002) that the training data set for each replicate, and holding out the burned between 1970 and 2002; thus our training data are other half to test the model’s predictions. In other words, likely biased. However, we are uncertain how this bias may MaxEnt produced 10 models using 10 randomly generated have affected our model. It is possible that the training data subsets of the 1970–2002 large-fire data, each consisting better represent large wildfires that were more difficult to of 802 points. Then each of these models was tested using suppress or contain because of inaccessibility owing to the subset of large-fire points held out (n = 801). During this geography or absence of roads. To address this issue, we process, we increased the regularization multiplier from also included a variable that represents distance from roads. its default of 1.0, which helps to prevent model overfitting, We assumed that wildfires were more apt to get bigger when by increments of 0.5 to minimize the difference between farther away from a road because it was more difficult to get the regularized training gain and test gain, while tryng to people and equipment into the area to fight the fire. maximize the test area under the curve (AUC) statistic and Spearman rank (Rs) correlation coefficient on our held-out Wildfire Suitability Modeling test data. These three statistics (gain, AUC, and Rs) are We chose the same modeling features in MaxEnt that we commonly used to measure the discriminative and predic- used for habitat modeling (linear, product, and hinge tive power of these sorts of models (Boyce et al. 2002, features) because this combination works well in fitting Fielding and Bell 1997, Hirzel et al. 2006). the environmental data to known or expected relationships The gain relates to how different the training or testing between the environmental variables and wildfires based data are from the background data. It is similar to “devi- on visual review of response curves generated during the ance” as used in generalized linear modeling (Phillips et al. modeling procedure. This combination is also a compro- 2006), and higher gains indicate larger differences between mise between using features that may be too restrictive for occurrence location environmental conditions and average complex environmental relationships (i.e., only linear) while background environmental conditions. The exponent of gain

70 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

produces the mean probability value of occurrence com- (this report), our final model product is the “average” model pared to random locations selected from the surrounding from our bootstrapped replicates. The predictive qualities of modeled landscape. Large differences between the regular- the “average” map can be better explained by the diagnostic ized training and testing gains indicates model overfitting. predicted versus expected (P/E) curve (fig. 4.6) (Hirzel et The AUC statistic is a measure of the model’s predic- al. 2006), and this curve allows users to better interpret the tive accuracy, and it was originally developed for evalua- modeled values. tions using presence and absence data, producing an index We also analyzed the importance of each environ- value from 0.5 to 1 with values close to 0.5 indicating poor mental variable and its relationship with large wildfire discrimination and a value of 1 indicating perfect predic- occurrence by running jackknifed models (Phillips et al. tions. The AUC values can be interpreted similarly to the 2006) for each of the 10 replicates. For each environmental traditional academic point system where values between variable, this jackknifing procedure produces a model that 0.9 and 1.0 indicate an excellent model (A), 0.8 to 0.9 is excludes the variable, and another model based on only that good (B), 0.7 to 0.8 is fair (C), 0.6 to 0.7 is poor (D), and variable. The gain and AUC model performance statistics AUC values between 0.5 and 0.6 represent failure (F), or from the jackknifed models then inform us on the relation- models that don’t predict much better than a random guess. ship and importance of each variable in explaining large- Examples of this interpretation in the field of niche-based wildfire occurrence in the area being modeled. species distribution models can be found in Araújo et al. (2005) and Randin et al. (2006). In our situation, MaxEnt Results uses 10,000 randomly selected background locations (map An average testing gain of 0.80 indicates that our model pixels) instead of true absence data, so it is not possible to predicted large-wildfire occurrence 2.2 times that expected achieve an AUC value of 1.0 (Wiley et al. 2003). However, by chance. The testing gain was also similar to the regular- interpretation is similar, with higher AUCs indicating better ized training gain of 0.77 indicating that our model was not model predictions (Phillips et al. 2006). Specific to our over-fit to the environmental data. The mean testing data case, AUC values represent the percentage of times a large AUC, based on 10 bootstrapped replicates, was 0.83, and wildfire location would have a higher wildfire suitability using independent test data from large wildfires from 2003 value than a randomly selected location from the modeling to 2009, the AUC was 0.78. The replicate mean predicted region. versus expected (P/E) curve (Hirzel et al. 2006) had an Rs = The Spearman rank correlation coefficient is a non- 1.0 (P < 0.001) and the test data P/E curve had an Rs = 0.987 parametric statistic that, in our situation, compares the (P < 0.001). The highest mean logistic probability for our ranks of large fire occurrence vs. area available to “binned” model was 0.90, which we converted into an integer value modeled prediction ranks (Boyce et al. 2002). A good model (90) for GIS mapping purposes by multiplying by a factor would predict an increasing ratio of the percentage of fire of 100. The threshold of 31 along this probability gradient occurrence to the percentage of the modeled landscape in marks where the predicted probability of large-wildfire oc- each model bin as the bin values increase, and an Rs of 1.0 currence is greater than what would be expected by chance indicates a strong positive correlation (Boyce et al. 2002). (fig. 4-6). One can use that threshold to define the owl’s The best model using the training data, and based on range in binary terms, where mapped values above this these statistics was produced using a regularization multi- threshold represent geographic areas that are more prone plier of 1.5. We then reran the same model, using the entire to large wildfire occurrence, based on our 32-year training training data set (n = 1,603) and conducted a final test of data timeframe, and areas below that threshold are not the model using 7 years of independent test data from large normally prone to large wildfires during that timeframe. fires that occurred between 2003 and 2009. Following the The strongest environmental variables were August same rationale and modeling approach used in chapter 3 maximum temperature, slope, and lightning ignition

71 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-6—Results of model bootstrapped replicates (based on fire data from 1970 to 2002) and independent testing (dashed-line based on fire data from 2003 to 2009) are shown in this predicted vs. expected curve (Hirzel et al. 2006). The curves indicate that the model performed well in both tests. The point at which the mean curve crosses the random frequency line (predicted/expected = 1) is used as the threshold modeled likelihood value (>32) for delineating the “fire-prone” areas of the binary map from the full gradient map. The gray-shaded area along the mean curve represents the 95-percent confidence intervals (95% CI) from the bootstrapped replicates.

72 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

density, explaining 76 percent of the geographical patterns covered with ponderosa pine forests, known for its depen- of large wildfires. Distance to roads contributed another 15 dency on wildfire. This, and recent occurrences of large percent, and, together, these four variables account for 91 wildfires in these areas (i.e., B&B Complex, Link Fire, percent of the information that relates to wildfire suitability Davis Lake Fire, and the Eyerly Fire) may point to potential in our model. Response curves (app. G) suggest suitability model limitations, which we discuss below. for large wildfires increases (in almost a logistic fashion) To begin with, our map represents suitability for what as August maximum temperature and slope increases. The we defined as “large” wildfires, and perhaps ones that are response curves for lightning ignition density and distance harder to suppress and contain, given the potential bias of to road variable response curves are quadratic in shape, our training data. The map does not represent a suitability showing sharp increases in suitability as the variable value gradient for all wildfire occurrences, nor behaviors, such increases, but then reaching a wide plateau and eventually as fire severity. Secondly, our map was trained with about decreasing (app. G). We suspect this decrease in suitability three decades of large wildfire data and therefore, repre- at the high end is related to elevation effects, which also sents the likelihood of large wildfire occurrence within exhibited a similar quadratic response curve. Extreme that specific timeframe. If we go further back in time, the distances from roads occurred in many wilderness areas fire history record within the owl’s range clearly shows the located along the Cascades crest, and these remote areas occurrence of large wildfires in the lower suitability areas tend to be at the highest elevations where late snowmelt (“bluer areas”) of our map (fig. 4-3), such as the Yacolt and produces cooler and moister conditions during the fire Columbia Fires of 1902, the Tillamook Fire of 1933, and the season. Likewise, lightning ignitions tend to be highest at more recent Oxbow Fire of 1966 (see fig. 3-16 in chapter 3). high elevations. Of the six model variables used, slope had As noted above, our model is based on climate and topo- the highest gain when modeled by itself. It also decreased graphic variables that have been relatively stable over the the gain the most when omitted from the model, and last century. The large wildfires that have occurred within therefore is an important variable in our model and appears the lower suitability areas of our map have been consistently to have the most information that is not present in the other associated with extreme weather (i.e., high winds) or heavy, environmental variables. contiguous, dry fuel (McClure 2005, Morris 1935), which could not be included in the model. We suspect a map Discussion characterizing long-term means of these extreme, episodic Four decades of history on large-wildfire occurrence fit well climatic events would be even more difficult to produce within our map of “wildfire suitability” gradients (fig. 4-7). than a rangewide fuel map. However, fire ecologists have The binary version of our map (fig. 4-8) has some distinct recently divided the range of the owl into five “fire regime similarities to previously mapped versions of “fire-prone” groups,”5 which represent a coarse spatial integration of fire areas in the owl’s range, especially the map by Agee and frequency and severity (Keane et al. 2002, Morgan et al. Edmonds (1992). But, it also has some distinct differences; 2001, Schmidt et al. 2002). Whereas fire regimes relate to most notably, it includes the considerable portions of the the frequency, severity, and spatial distribution of historical western Cascades of Oregon, and it excludes large areas of wildfire in the ecosystem (Rollins et al. 2002), our map the eastern Cascades that are commonly shown on previous sheds light only on the latter of these three characteristics. map versions (fig. 4-1). Based on our map, only the northern However, it still shows spatial similarities to the fire regime half of the Washington Eastern Cascades physiographic group map, and, in particular, the lower suitability areas province has substantial land area that appears suitable for complement Fire Regime Group V, which represents infre- large-wildfire occurrence. South of this, our map indicates quent fires (>200-year intervals) and mostly occurs in the a patchy distribution of high-suitability areas along the 5 LANDFIRE data products and their descriptions are available eastern Cascades; yet, much of this area historically was online at http://www.landfire.gov/products_overview.php.

73 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-7—The full gradient version of the wildfire suitability model showing locations of large wildfires used to train the model (left) and locations of large wildfires that occurred after 2002 (right) that served as our independent testing data.

74 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 4-8—The binary version of the wildfire suitability model showing locations of large wildfires used to train the model (left) and locations of large wildfires that occurred after 2002 (right) that served as our independent testing data.

75 GENERAL TECHNICAL REPORT PNW-GTR-850

coastal zones and highest elevations of the mountain ranges. effect on the distribution, although that evidence is indirect These areas have been defined as incurring infrequent as distribution on the landscape was not the focus of any of wildfires that, when they do happen, tend to be extremely these studies. Because we have almost no data on how the large and severe (Agee 1993, 1998; Morgan et al. 2001; Noss distribution of large fires might have differed in the absence et al. 2006). of suppression actions within the study area, we cannot On the other end of the fire regime group spectrum, characterize any model bias in that regard. wildfires were more frequent (<35-year intervals) and less Using forest health protection aerial survey data from severe, maintaining open forests, or a mosaic of different- 1983 to 2008 (USDA 2008), spatial patterns of recent aged forest seral stages (Hann and Strohm 2003). On our western spruce budworm (Choristoneura occidentalis) and map, portions of the Cascades east of the high-elevation mountain pine beetle (Dendroctonus ponderosae) outbreaks crest, where ponderosa pine forests historically dominated, become apparent (fig. 4-9). In 1983, the spruce budworm fit this description. These pine forests were once dependent began expanding its distribution in the eastern Cascades of on frequent surface fires that burned heterogeneously Oregon, spreading northward into Washington. It mostly through the landscape, creating open, parklike distribu- ran its course in the eastern Oregon Cascades by 1993, and tions of trees that were often clumped into small groups then became more active in the southern portions of the (Agee 1994, Graham and Jain 2005). Historically, wildfires Washington Eastern Cascades province. The increased fuel in ponderosa pine forests were relatively easy to contain loads created by severe insect outbreaks certainly increase (Munger 1917). Fire suppression and exclusion in ponderosa suitability for large wildfires, especially if the fuels are pine forests produced changes that have been well docu- concentrated in a contiguous fashion. In general, the spatial mented by scientists since the 1990s (Agee 1990, Deeming pattern of concentrated spruce budworm outbreaks cor- 1990, Kauffman 1990, Mitchell 1990, Mutch et al. 1993, respond highly with the B&B Complex and Link Fires from Wickman 1992). The lack of natural wildfires allows 2003 (fig. 4-9), and also the Lake George, Puzzle, and Black understory development of shade-tolerant vegetation that Crater Fires from 2006, and the loss of owl habitat in these produces resource-stressed stands, making them more areas has largely been attributed to this spruce budworm susceptible to insects and disease. This, in turn, leads to epidemic and its contribution to the wildfire’s size and weakened or dead trees, generating fuel loadings that are severity (Courtney et al. 2004). In addition, the Davis Fire unnaturally heavy and also contiguous over large areas of 2003 occurred in a concentrated area of recent mountain (Hessburg et al. 2005). The understory development pine beetle outbreaks, and we consider it is likely these also creates ladder fuels that can lead to crown fires, and episodic insect infections added to the suitability of those this fuel combination produces conditions ripe for large specific areas to support these large wildfires. wildfires (Hessburg et al. 2005). Today, these areas have Agee (1993) pointed out that fire regimes are dependent developed fuel characteristics that support larger, and more on the interaction of all parts of the fire behavior triangle— severe, wildfires (Hessburg et al. 2005), and the recent weather, topography, and fuel. Parisien and Moritz (2009) wildfires in central eastern Oregon Cascades have been described the fire regime triangle as the interaction of larger than those of historical records (Eckert et al. 2008). climate, ignitions, and vegetation. Our map spans both the To what extent fire suppression may have biased our spatial and temporal scales that these triangles represent map is uncertain. We suspect fire suppression has likely (see fig. 1 in Parisian and Moritz 2009) and appears to affected the frequency of large wildfires, but it is much less reasonably reflect the last four decades of wildfire history clear that it has affected the distribution of large wildfires within the range of the owl. However, we suspect that ad- on the landscape. Studies of large wildfires in the large ditional spatial information on long-term means of episodic wilderness areas of the southwest and northern Rockies climatic events or insect outbreaks would likely increase its (Rollins et al. 2002, 2004) suggest there has not been an accuracy.

76 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure 4-9—Episodic outbreaks of insects in the Oregon Cascades provinces beginning in the 1980s, shown above, may help explain why portions of the eastern Cascades of Oregon have experienced recent large wildfires in areas where the wildfire suitability map indicates lower wildfire suitability (blue indicates lower amounts in the maps above, such as wildfire suitability or years of insect detection, whereas red indicates higher amounts). Large wildfires from 1970 to 2009 are shown as black cross-hatched polygons.

Perhaps one of the most compelling validations of our shorter fire-return intervals (Skinner and Chang 1996, wildfire suitability map is the relationship of the distribu- Taylor and Skinner 1998). These species are members of the tions of three fire-dependent pine species (Little 1971, USDI “fire-resistant” group of pines (McCune 1988) that evolved 1999) with our binary characterization of wildfire suitability in fire-prone environments and developed characteristics (fig. 4-10). As a group in general, pines are associated with like thick bark to insulate the cambium and long needles forests where wildfire is an integral part of the environment to insulate buds from the heat of wildfires. Using a map (Fonda 2001). In the range of the owl, sugar pine (Pinus comparison technique6 (Visser and de Nijs 2006), we lambertiana Dougl.), Jeffrey pine (P. jeffreyi Grev. & 6 This analysis was performed by using the Map Comparison Kit Balf.), and ponderosa pine (P. ponderosa P. & C. Lawson) software (version 3.2) (Netherlands Environmental Assessment are common associates with fire-prone ecosystems having Agency) developed by the Research Institute for Knowledge Systems, and available online at http://www.riks.nl/mck/.

77 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 4-10—Fire-resistant pine distribution maps that were delineated in 1971 (Little 1971, USDI 1999) overlaid on the wildfire suitability binary map.

found that their combined geographic distributions (Little Summary 1971) coincide (cell to cell) moderately well (kappa = 0.46, Our goal was to identify landscape-scale areas within the KLOC = 0.76, KHISTO = 0.60) with the fire-prone areas of owl’s range where large wildfires are more probable over the binary version of our model. We believe the historical time using factors that are mostly spatially and temporally distributions of these fire-resistant conifers provide further stable. The use of topographic and climate variables that evidence that our binary map effectively identifies portions summarized weather patterns over multiple years (1970– of the owl’s range that wildfire regularly “inhabits.” 2000) resulted in a map that we believe met this goal, as Countryman (1966) described “fire environment” as evidenced by the map’s moderate to good correlations (AUC the complex of fuel, topographic, and weather (air mass) of 0.78 to 0.83 and Rs ≥ 0.987) with large wildfire locations factors that influences the inception, growth, and behavior that post-date the wildfire data used to train the model as of fire. He realized that it was a “pattern phenomena,” and well as historical distribution maps of fire-dependent pine advised that its pattern “must be considered in order to species (fig. 4-10). A binary classification of our map (based understand and predict a fire’s behavior.” Our map of wild- on the threshold where the map predicts large wildfires fire suitability is essentially a modeling application of the more often than would be expected by chance) provides concept Countryman first described over 40 years ago and a less arbitrary way to identify “fire-prone” areas of the is well-validated by almost 40 years of large wildfire data. northern spotted owl’s range that normally experience large wildfires.

78 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

With this knowledge, we can overlay our wildfire suit- source of habitat loss and future recruitment on federal ability map on the current habitat suitability map produced lands in certain parts of the owl’s range. Fortunately, our in chapter 3, to confirm that the physiographic province capabilities to map owl habitat suitability, wildfire effects with the most owl habitat in fire-prone landscapes is the on vegetation, and wildfire suitability are improving; California Klamath province (fig. 4-11). The next highest informing us better on what the habitat effects might be and province is the Oregon Western Cascades province. How- where this interaction is most likely to happen. ever, the recent occurrence and trends of insect outbreaks A limitation of our map is that, by itself, it does not pro- in the eastern Cascades needs to be considered as well. The vide information on where within the range, large wildfires effects of past management practices combined with these may occur as a result of atypical or unusual, infrequent outbreaks have probably increased the suitability for large conditions or events such as extreme fire weather (Bessie wildfires of areas that otherwise have underlying physical and Johnson 1995; Westerling et al. 2003, 2006), fuel condi- and climatic factors that are not suitable. If this is the case, tions, or a combination of the two. There are other tools our results suggest that once the current fuel build-ups in available to monitor and track those conditions. However, the eastern Cascade provinces are reduced to more natural our map can be used in conjunction with this ancillary data, levels, the occurrence of large wildfires in that area should such as insect outbreak maps, to better inform us on where decline. the next large wildfires might happen. The effects of wildfire on owl biology are difficult Finally, the inclusion of climate variables that sum- to assess and will likely remain a source of uncertainty marize fire weather in our model may give us the ability to (Courtney et al. 2004) for some time. Yet, the latest esti- explore climate change scenarios (Carroll 2010) and what mates of wildfire’s effect on current and future owl habitat, effect they may have on patterns of wildfire suitability in as displayed in chapter 3, indicate wildfires are the major the future.

Figure 4-11—Fire-prone spotted owl nesting/roosting habitat, both reserved and nonreserved, by physiographic province, 2006/07. The majority of the fire-prone habitat occurs within the Klamath provinces, and the southern portions of the Oregon Western Cascades. Over half is in reserved land use allocations.

79 GENERAL TECHNICAL REPORT PNW-GTR-850

References Beyer, H.L. 2009. Hawth’s analysis tools (version 3.27) for Agee, J.K. 1990. The historical role of fire in Pacific ArcGIS. http://www.spatialecology.com/htools. Northwest forests. In: Walstad, J.D.; Radosevich, S.R.; (4 January 2011). Sandberg, D.V., eds. Natural and prescribed fire in Bingham, B.; Noon, B.R. 1997. Mitigation of habitat Pacific Northwest Forests. Corvallis, OR: Oregon State take: application to habitat conservation planning. University Press: 25–38. Conservation Biology. 11(1): 127–139. Agee, J.K. 1993. Fire ecology of Pacific Northwest forests. Bond, M.L.; Lee, D.E.; Siegel, R.B.; Ward, J.P., Jr. 2009. Washington, DC: Island Press. 493 p. Habitat use and selection by California spotted owls in Agee, J.K. 1994. Fire and weather disturbances in a postfire landscape. Journal of Wildlife Management. terrestrial ecosystems of the eastern Cascades. Gen. 73(7): 1116–1124. Tech. Rep. PNW-GTR-320. Portland, OR: U.S. Bond, W.J.; Keeley, J.E. 2005. Fire as a global Department of Agriculture, Forest Service, Pacific “herbivore”: the ecology and evolution of flammable Northwest Research Station. 52 p. ecosystems. Trends in Ecology and Evolution. 20(7): Agee, J.K. 1998. The landscape ecology of western forest 387–394. fire regimes. Northwest Science. 72(Spec. issue): 24–34. Boyce, M.S.; Vernier, P.R.; Nielsen, S.E.; Schmiegelow, Agee, J.K.; Edmonds, R.L. 1992. Appendix F—forest F.K.A. 2002. Evaluating resource selection functions. protection guidelines for the northern spotted owl. In: Ecological Modelling. 157: 281–300. Final draft recovery plan for the northern spotted owl: Brown, A.A.; Davis, K.P. 1973. Forest fire control and use. 419–480. Unpublished report. On file with: U.S. Fish and New York, NY: McGraw-Hill Book Company. 686 p. Wildlife Service, 2600 SE 98th Ave., Suite 100, Portland, Brown, T.J.; Hall, B.L.; Mohrle, C.R.; Reinbold, OR 97266. H.J. 2002. Coarse assessment of federal wildland Albini, F. 1976. Estimating wildfire behavior and effects. fire occurrence data: report for the National Wildfire Gen. Tech. Rep. INT-30. Ogden, UT: U.S. Department Coordinating Group. Reno, NV: CEFA Report 02-04. of Agriculture, Forest Service, Intermountain Forest and Program for Climate, Ecosystem and Fire Applications, Range Experiment Station. 92 p. Desert Research Institute, Division of Atmospheric Albini, F.A.; Latham, D.J.; Baughman, R.G. 1982. Sciences. 31 p. Estimating upslope convective wind speeds for Carroll, C. 2010. Role of climatic niche models in focal- predicting wildland fire behavior. Res. Pap. INT-257. species-based conservation planning: assessing potential Ogden, UT: U.S. Department of Agriculture, Forest effects of climate change on northern spotted owl in the Service, Intermountain Forest and Range Experiment Pacific Northwest, USA. Biological Conservation. 143(6): Station. 19 p. 1432–1437. Araújo, M.B.; Pearson, R.G.; Thuillers, W.; Erhard, M. Clark, D.A. 2007. Demography and habitat selection 2005. Validation of species-climate impact models under of northern spotted owls in post-fire landscapes of climate change. Global Change Biology. 11: 1504–1513. southwestern Oregon. Corvallis, OR: Oregon State Bessie, W.C.; Johnson, E.A. 1995. The relative importance University. 218 p. M.S. thesis. of fuels and weather on fire behavior in subalpine forests. Countryman, C.M. 1964. Mass fires and fire behavior. Ecology. 76: 747–762. Res. Rep. PSW-19. Berkley, CA: U.S. Department of Agriculture, Forest Service, Pacific Southwest Forest and Range Experiment Station. 53 p.

80 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Countryman, C.M. 1966. The concept of fire environment. Elith, J.; Phillips, S.J.; Hastie, T.; Dudík, M.; Chee, Fire Control Notes. Berkley, CA: U.S. Department of Y.E.; Yates, C.J. 2011. A statistical explanation of Agriculture, Forest Service, Pacific Southwest Forest and MaxEnt for ecologists. Diversity and Distributions. 17: Range Experiment Station. 27(4): 8–10. 43–57. Courtney, S.P.; Blakesley, J.A.; Bigley, R.E.; Cody, Fielding, A.H.; Bell, J.F. 1997. A review of methods for M.L.; Dumbacher, J.P.; Fleischer, R.C.; Franklin, assessment of prediction errors in conservation presence/ A.B.; Franklin, J.F.; Gutiérrez, R.J.; Marzluff, J.M.; absence models. Environmental Conservation. 24: Sztukowski, L. 2004. Scientific evaluation of the status 38–49. of the northern spotted owl. Portland, OR: Sustainable Fonda, R.W. 2001. Burning characteristics of needles from Ecosystem Institute. 348 p. + appendixes. eight pine species. Forest Science. 47: 390–396. Cui, W.; Perera, A.H. 2008. What do we know about Forest Ecosystem Management Assessment Team forest fire size distribution, and why is this knowledge [FEMAT]. 1993. Forest ecosystem management: an useful for forest management? International Journal of ecological, economic, and social assessment. Portland, Wildland Fire. 17: 234–244. OR: U.S. Department of Agriculture; U.S. Department of Davis, R.; Lint, J. 2005. Habitat status and trend. In: the Interior [and others]. [Irregular pagination]. Lint, J., tech. coord. Northwest Forest Plan—the first Franklin, A.B.; Anderson, D.R.; Gutiérrez, R.J.; 10 years (1994–2003): status and trends of northern Burnham, K.P. 2000. Climate, habitat quality, spotted owl populations and habitat. Gen. Tech. Rep. and fitness in northern spotted owl populations in PNW-GTR-648. Portland, OR: U.S. Department of northwestern California. Ecological Monographs. 70(4): Agriculture, Forest Service, Pacific Northwest Research 539–590. Station: 21–82. Franklin, J.F.; Dyrness, C.T. 1973. Natural vegetation Deeming, J.E. 1990. Effects of prescribed fire on wildfire of Oregon and Washington. Gen. Tech. Rep. PNW-8. occurrence and severity. In: Walstad, J.D.; Radosevich, Portland, OR: U.S. Department of Agriculture, Forest S.R.; Sandberg, D.V., eds. Natural and prescribed fire in Service, Pacific Northwest Forest and Range Experiment Pacific Northwest forests. Corvallis, OR: Oregon State Station. 417 p. University Press: 95–104. Garfin, J.E.; Morehouse, B., eds. 2001. Fire and climate Deeming, J.E.; Burgan, R.E.; Cohen, J.D. 1977. The workshop proceedings. Tucson, AZ: University of national fire danger rating system. Gen. Tech. Rep. INT- Arizona, Institute for the Study of Planet Earth, Climate 39. Ogden, UT: U.S. Department of Agriculture, Forest Assessment for the Southwest. 75 p. Service, Intermountain Forest and Range Experiment Gisborne, H.T. 1936. Measuring fire weather and forest Station. 63 p. inflammability. Circular No. 398. Washington, DC: U.S. Eckert, B.E.; Walstad, J.D.; Tappeiner, J.C., II. 2008. Department of Agriculture. 59 p. An illustrated guide to fire in central Oregon forests. Graham, R.T.; Jain, T.B. 2005. Ponderosa pine eco- Contributions in Education and Outreach No. 1. systems. In: Ritchie, M.W.; Maguire, D.A.; Youngblood, Corvallis, OR: Forest Research Laboratory, Oregon State A., tech. cords. 2005. Proceedings of the symposium on University. 40 p. ponderosa pine: issues, trends, and management. Gen. Eidenshink, J.; Schwind, B.; Brewer, K.; Zhu, Z.L.; Tech. Rep. PSW-GTR-198. Albany, CA: U.S. Department Quayle, B.; Howard, S. 2007. A project for monitoring of Agriculture, Forest Service, Pacific Southwest trends in burn severity. Fire Ecology. 3: 3–21. Research Station: 1–32.

81 GENERAL TECHNICAL REPORT PNW-GTR-850

Grinnell, J. 1917. Field tests of theories concerning Kauffman, J.B. 1990. Ecological relationships of distributional control. The American Naturalist. 51(602): vegetation and fire in Pacific Northwest forests. In: 115–128. Walstad, J.D.; Radosevich, S.R.; Sandberg, D.V., eds. Natural and prescribed fire in Pacific Northwest forests. Guisan, A.; Zimmermann, N.E. 2000. Predictive habitat Corvallis, OR: Oregon State University Press: 39–52. distribution models in ecology. Ecological Modelling. 135: 147–186. Kaufmann, J.H. 1962. Ecology and social behavior of the coati, Nasua nirica on Barro Colorado Island Panama. Hann, W.J.; Strohm, D.J. 2003. Fire regime condition University of California Publications in Zoology. 60: class and associated data for fire and fuels planning: 95–222. methods and applications. In: Omi, P.N.; Joyce, L.A., eds. Fire, fuel treatments and ecological restoration: Keane, R.E.; Parsons, R.; Hessburg, P. 2002. Estimating conference proceedings. Proceedings RMRS-P-29. Fort historical range and variation of landscape patch Collins, CO: U.S. Department of Agriculture, Forest dynamics: limitations of the simulation approach. Service, Rocky Mountain Research Station: 397–433. Ecological Modeling. 151: 29–49. Hanson, C.T.; Odion, D.C.; Dellasala, D.A.; Baker, B.L. Komarek, E.V. 1967. The nature of lightning fire. In: 2009. Overestimation of fire risk in the northern spotted Komarek, E.V., ed. Proceedings, 7th Tall Timbers fire owl recovery plan. Conservation Biology. 23: 1314–1319. ecology conference. Tallahassee, FL: Tall Timbers Research Station: 5–41. Hayes, G.L. 1941. Influence of altitude and aspect on daily variations in factors of forest-fire danger. Circular No. Krawchuk, M.A.; Moritz, M.A.; Parisien, M.A.; Van 591. Washington, DC: U.S. Department of Agriculture. Dorn, J.; Hayhoe, K. 2009. Global pyrogeography: the 38 p. current and future distribution of wildfire. Public Library of Science One. 4(4):e5102. DOI: 10.1371/journal. Healey, S.P.; Cohen, W.B.; Spies, T.A.; Moeur, M.; pone.0005102. Pflugmacher, D.; Whitley, M.G.; Lefsky, M. 2008. The relative impact of harvest and fire upon landscape-level Lint, J.; Noon, B.; Anthony, R.; Forsman, E.; Raphael, dynamics of older forests: lessons from the Northwest M.; Collopy, M.; Starkey, E. 1999. Northern spotted Forest Plan. Ecosystems. 11: 1106–1119. owl effectiveness monitoring plan for the Northwest Forest Plan. Gen. Tech. Rep. PNW-GTR-440. Portland, Hessburg, P.F.; Agee, J.K.; Franklin, J.F. 2005. Dry OR: U.S. Department of Agriculture, Forest Service, forests and wildland fires of the inland Northwest USA: Pacific Northwest Research Station. 43 p. contrasting the landscape ecology of the pre-settlement and modern eras. Forest Ecology and Management. 211: Little, E.L., Jr. 1971. Atlas of United States trees, 117–139. volume 1, conifers and important hardwoods. U.S. Department of Agriculture Misc. Publ. 1146. 9 p. Hirzel, A.H.; Le Lay, G. 2008. Habitat suitability +200 maps. modelling and niche theory. Journal of Applied Ecology. 45: 1372–1381. McClure, R. 2005. Washington’s “awful conflagration”— The Yacolt Fire of 1902. Fire Management Today. 65(1): Hirzel, A.H.; Le Lay, G.; Helfer, V.; Randin, C.; Guisan, 24–27. A. 2006. Evaluating the ability of habitat suitability models to predict species presences. Ecological McCune, B. 1988. Ecological diversity in North American Modelling. 199: 142–152. pines. American Journal of Botany. 75(3): 353–368. Hutchinson, G.E. 1957. Concluding remarks. Cold Spring Harbor Symposium. Quantitative Biology. 22: 415–427.

82 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Miller, J.D.; Safford, H.D.; Crimmins, M.A.; Thode, Parisien, M.; Moritz, M.A. 2009. Environmental controls A.E. 2009. Quantitative evidence for increasing forest on the distribution of wildfire at multiple spatial scales. fire severity in the Sierra Nevada and southern Cascade Ecological Monographs. 79(1): 127–154. Mountains, California and Nevada, USA. Ecosystems. Phillips, S.J.; Anderson, R.P.; Shapire, R.E. 2006. 12: 16–32. Maximum entropy modeling of species geographic Mitchell, R.G. 1990. Effects of prescribed fire on insect distributions. Ecological Modelling. 190: 231–259. pests. In: Walstad, J.D.; Radosevich, S.R.; Sandberg, Phillips, S.; Dudík, M. 2008. Modeling of species D.V., eds. Natural and prescribed fire in Pacific North- distributions with MaxEnt: new extensions and a west forests. Corvallis, OR: Oregon State University comprehensive evaluation. Ecography. 31: 161–175. Press: 111–116. Pierce, K.B.; Lookingbill, T.R.; Urban, D.L. 2005. Morgan, P.; Hardy, C.C.; Swetnam, T.W.; Rollins, M.G.; A simple method for estimating potential relative Long, D.G. 2001. Mapping fire regimes across time and radiation (PRR) for landscape-scale vegetation analysis. space: understanding coarse and fine-scale fire patterns. Landscape Ecology. 20: 137–147. International Journal of Wildland Fire. 10: 329–342. Potter, B.E. 1996. Atmospheric properties associated with Morris, W.G. 1934. Lightning storms and fires. Portland, large wildfires. International Journal of Wildland Fire. OR: U.S. Department of Agriculture, Forest Service, 6(2): 71–76. Pacific Northwest Forest and Range Experiment Station. Preisler, H.K.; Westerling, A.L. 2007. Statistical model 27 p. for forecasting monthly large wildfire events in Western Morris, W.G. 1935. The details of the Tillamook fire United States. Journal of Applied Meteorology and from its origin to the salvage of the killed trees. 26 p. Climatology. 46(7): 1020–1030. Unpublished report. On file with: National Archives Pyne, S.J. 2001. Fire: a brief history. Seattle, WA: Records Admin., Pacific Alaska Region, 6125 Sand Point University of Washington Press. 209 p. Way NE, Seattle, WA 98115-7999. Pyne, S.J. 2004. Tending fire: coping with America’s Munger, T.T. 1917. Western yellow pine in Oregon. wildland fires. Washington, DC: Island Press. 256 p. Bulletin 418. Washington, DC: U.S. Department of Agriculture. 48 p. Randin, C.F.; Dirnbo, T.; Dullinger, S.; Zimmermann, N.E.; Zappa, M.; Guisan, A. 2006. Are niche-based Mutch, R.W.; Arno, S.F.; Brown, J.K.; Carlson, species distribution models transferable in space? Journal C.E. 1993. Forest health in the Blue Mountains: a of Biogeography. 33: 1689–1703. management strategy for fire-adapted ecosystems. Gen. Tech. Rep. PNW-310. Portland, OR: U.S. Department of Rapp, V. 2005. Conserving old forest in landscapes Agriculture, Forest Service, Pacific Northwest Forest and shaped by fire. Science Update 11. Portland, OR: U.S. Range Experiment Station. 14 p. Department of Agriculture, Forest Service, Pacific Northwest Research Station. 11 p. Noss, R.F.; Franklin, J.F.; Baker, W.L.; Schoennagel, T.; Moyle, P.B. 2006. Managing fire-prone forests in Rollins, M.G.; Keane, R.E.; Parsons, R.A. 2004. the Western United States. Frontiers in Ecology and the Mapping fuels and fire regimes using remote sensing, Environment. 4: 481–487. ecosystem simulation, and gradient modeling. Ecological Applications. 14(1): 75–95. Oregon Climate Service. 2008. Parameter-elevation re- gressions on independent slopes models (PRISM). http:// www.ocs.oregonstate.edu/prism. (21 December 2010).

83 GENERAL TECHNICAL REPORT PNW-GTR-850

Rollins, M.G.; Morgan, P.; Swetnam, T. 2002. Scott, J.H.; Reinhardt, E.D. 2001. Assessing crown fire Landscape-scale controls over 20th century fire potential by linking models of surface and crown fire occurrence in two large Rocky Mountain (USA) behavior. Res. Pap. RMRSRP- 29. Fort Collins, CO: wilderness areas. Landscape Ecology. 17: 539–557. U.S. Department of Agriculture, Forest Service, Rocky Rorig, M.L.; Ferguson, S.A. 1999. Characteristics of Mountain Research Station. 59 p. lightning and wildland fire ignition in the Pacific Sensenig, T.S. 2002. Development, fire history and current Northwest. Journal of Applied Meteorology. 38: and past growth, of old-growth and young-growth forest 1565–1575. stands in the Cascade, Siskyou and mid-Coast Mountains Rothermel, R.C. 1983. How to predict the spread and of southwestern Oregon. Corvallis, OR: Oregon State intensity of forest and range fires. Res. Pap. INT-143. University. 180 p. Ph.D. dissertation. Ogden, UT: U.S. Department of Agriculture, Forest Skinner, C.N.; Chang, C. 1996. Fire regimes, past and Service, Intermountain Forest and Range Experiment present. In: Sierra Nevada Ecosystem Project: final Station. 161 p. report to Congress. Volume II, assessments and scientific Rothermel, R.C. 1991. Predicting behavior and size of basis for management options. Davis, CA: University of crown fires in the northern Rocky Mountains. Res. Pap. California, Centers for Water and Wildland Resources: INT-438. Ogden, UT: U.S. Department of Agriculture, 1041–1069. http://ceres.ca.gov/snep/pubs/web/PDF/ Forest Service, Intermountain Research Station. 46 p. VII_C38.PDF. (18 December 2010). Samuel, M.D.; Pierce, D.J.; Garton, E.O. 1985. Spies, T.A.; Hemstrom, M.A.; Youngblood, A.; Hummel, Identifying areas of concentrated use within the home S. 2006. Conserving old-growth forest diversity in range. Journal of Animal Ecology. 54(3): 711–719. disturbance-prone landscapes. Conservation Biology. 20: 351–362. Schmidt, K.M.; Menakis, J.P.; Hardy, C.C.; Hann, W.J.; Bunnell, D.L. 2002. Development of coarse-scale spatial Stocks, B.J.; Mason, J.A.; Todd, J.B.; Bosch, E.M.; data for wildland fire and fuel management. Gen. Tech. Wotton, B.M.; Amiro, B.D.; Flannigan, M.D.; Hirsh, Rep. RMRS-GTR-87. Fort Collins, CO: U.S. Department K.G.; Logan, K.A.; Martell, D.L.; Skinner, W.R. of Agriculture, Forest Service, Rocky Mountain 2003. Large forest fires in Canada, 1959–1997. Journal of Research Station. 41 p. Geophysical Research. 108(D1, 8149): 1–12. Schroeder, M.J.; Buck, C.C. 1970. Fire weather: a guide Stratton, R.D. 2006. Guidance on spatial wildland fire for application of meteorological information to forest analysis: models, tools, and techniques. Gen. Tech. Rep. fire control operations. Agric. Handb. 360. Washington, RMRS-GTR-183. Fort Collins, CO: U.S. Department of DC: U.S. Department of Agriculture, Forest Service. Agriculture, Forest Service, Rocky Mountain Research 229 p. Station: 15 p. Schwind, B., comp. 2008. Monitoring trends in burn Sugihara, N.G.; van Wagtendonk, J.W.; Shaffer, severity: report on the Pacific Northwest and Pacific K.E.; Fites-Kaufamn, J.; Thode, A.E. 2006. Fire in Southwest fires (1984 to 2005). Unpublished report. California’s ecosystems. Berkley, CA: University of On file with: USDA Forest Service, Remote Sensing California Press. 596 p. Applications Center (RSAC), U.S. Geological Survey, Taylor, A.H.; Skinner, C.N. 1998. Fire history and Center for Earth Resources Observations and Science landscape dynamics in a late-successional reserve, (EROS). http://mtbs.gov/reports/MTBS_pnw-psw_final. Klamath Mountains, California USA. Forest Ecology pdf. (4 January 2011). and Management. 111: 285–301.

84 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Thomas, J.W.; Forsman, E.D.; Lint, J.B.; Meslow, E.C.; U.S. Department of the Interior, Geological Survey Noon, B.R.; Verner, J. 1990. A conservation strategy [USDI]. 1999. Digital representation of “Atlas of United for the northern spotted owl: a report of the Interagency States Trees” by Elbert L. Little, Jr,. USDI Geological Scientific Committee to address the conservation of the Survey, Earth Surface Processes. http://esp.cr.usgs.gov/ northern spotted owl. Portland, OR: U.S. Department data/atlas/little/. (4 January 2011). of Agriculture, Forest Service; U.S. Department of the Van Wagner, C.E. 1977. Conditions for the start and spread Interior, Bureau of Land Management, Fish and Wildlife of crown fire. Canadian Journal of Forest Research. 7: Service, National Park Service. 427 p. 23–34. U.S. Department of Agriculture, Forest Service [USDA]. Visser, H.; de Nijs, T. 2006. The map comparison kit. 2008. Forest health protection aerial survey data. http:// Environmental Modelling and Software. 21(3): 346–358. www.fs.fed.us/r6/nr/fid/as/index.shtml. (4 January 2011). Westerling, A.L.; Bryant, B.P. 2008. Climate change and U.S. Department of the Interior, Fish and Wildlife wildfire in California. Climatic Change. 87: 231–249. Service [USDI]. 1990. Endangered and threatened wildlife and plants: determination of threatened status Westerling, A.L.; Gershunov, A.; Brown, T.J. 2003. for the northern spotted owl. Federal Register 55: Climate and wildfire in the Western United States. 26114–26194. Bulletin of the American Meteorological Society. 84: 595–604. U.S. Department of the Interior, Fish and Wildlife Service [USDI]. 1992. Final draft recovery plan for the Westerling, A.L.; Hidalgo, H.G.; Cayan, D.R.; Swetnam, northern spotted owl. Unpublished report. On file with: T.W. 2006. Warming and earlier spring increase Western U.S. Fish and Wildlife Service, 2600 SE 98th Ave., Suite U.S. forest fire activity. Science. 313: 940–943. 100, Portland, OR 97266. Wickman, B.E. 1992. Forest health in the Blue Mountains: U.S. Department of the Interior, Fish and Wildlife the influence of insects and diseases. Gen. Tech. Rep. Service [USDI]. 2008. Final recovery plan for the PNW-GTR-295. Portland, OR: U.S. Department of northern spotted owl (Strix occidentalis caurina). 142 p. Agriculture, Forest Service, Pacific Northwest Research Unpublished report. On file with: U.S. Department of the Station. 15 p. (Quigley, Thomas M., ed.; Forest health in Interior, Fish and Wildlife Service, 2600 SE 98th Ave., the Blue Mountains science perspectives). Suite 100, Portland, OR 97266. Wiley, E.O.; McNyset, K.M.; Peterson, A.T.; Robins, U.S. Department of the Interior, Fish and Wildlife C.R.; Stewart, A.M. 2003. Niche modeling and Service; U.S. Department of the Interior, Bureau of geographic range predictions in the marine environment Land Management; U.S. Department of Agriculture, using a machine learning algorithm. Oceanography. 16: Forest Service [USDI and USDA]. 2008. Methodology 120–127. for estimating the number of northern spotted owls affected by proposed federal actions (version 2.0). Unpublished report. On file with: U.S. Fish and Wildlife Service, 2600 SE 98th Ave., Suite 100, Portland, OR 97266.

85 GENERAL TECHNICAL REPORT PNW-GTR-850 Ray Davis

86 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Chapter 5: Emerging Issues, Related Research, and Research Needs Katie M. Dugger and Raymond J. Davis Emerging Issues barred owls on spotted owl territories is associated with The 10-year report on the status and trends of northern decreased site occupancy by spotted owls and changes in spotted owl (Strix occidentalis caurina) populations and extinction and colonization rates (Dugger et al. 2008, 2009; habitat recognized that the conservation and recovery of the Kroll et al. 2010; Olson et al. 2005). The strongest associa- owl is not solely related to the amount and quality of habitat tion is between detections of barred owls and increased across its range (Lint 2005). Other factors including interac- extinction rates across the entire range of the spotted owl, tions with prey and prey biology, competition with barred and decreased colonization rates have been reported for owls (S. varia), and the emergence of West Nile virus in some study areas as well (Dugger et al. 2008, 2009; Kroll et the Pacific Northwest were noted as emerging issues (Lint al. 2010; Olson et al. 2005). The most recent meta-analysis 2005). The potential threat of West Nile virus infections to of spotted owl population dynamics reports a clear negative spotted owl populations has not been realized, despite early association between barred owl presence and spotted owl evidence that owls in the wild were susceptible to natural survival (Forsman et al. 2011; chapter 2, this report). Effects infection (Fitzgerald et al. 2003). For unknown reasons, on fecundity are less apparent, but declines in spotted owl outbreaks of the West Nile virus in spotted owls that were recruitment on four demographic study areas (Olympic, anticipated 5 years ago have not occurred, although the H.J. Andrews, Coast Ranges, Tyee) in association with virus is present throughout the owls range (Franklin 2010). barred owl presence has been reported (Glenn et al. 2010). Although West Nile virus has not developed into as much Thus, researchers continue to compile negative associations of a threat to owl populations as predicted previously between barred owl presence and spotted owl vital rates (Lint 2005), documentation of the negative association strengthening the evidence that barred owls are negatively between the invasive barred owl and spotted owl vital rates affecting spotted owl demography. has con-tinued over the last 5 years (Anthony et al. 2006, Climate change is another emerging issue that may Dugger et al. 2008, Forsman et al. 2011, Glenn et al. 2010, affect spotted owl habitat, populations, and the functionality Kroll et al. 2010). The barred owl is now found at signifi- of the network of reserved land use allocations across the cant densities throughout the entire range of the northern owl’s range (Carroll 2010, Carroll et al. 2009, Glenn et al. spotted owl (Livezey 2009), and the range expansion of this 2010, Spies et al. 2010). Forest Service research objectives species constitutes a significant threat to northern spotted include developing projections for changes in fire regimes owl persistence, which was not evident when the spotted and shifts in habitat distributions because altered forest owl was first listed (Courtney et al. 2004). The proportion structures with increased threats from wildfire and insect of spotted owl territories where barred owls have been and disease outbreaks are anticipated in association with detected has increased steadily since the early 1990s in the predicted climate change (USDA 2009). Rate of change in eight effectiveness monitoring areas administered under spotted owl population was negatively associated with hot, the Northwest Forest Plan (the Plan) (fig. 5.1) along with dry growing seasons and wet, stormy winters (Glenn et al. st increased evidence of negative interactions, presumably 2010). Climate models for the first half of the 21 century owing to competition or interference between the two predict warmer, wetter winters and hotter, drier summers, species (Dugger et al. 2008, 2009; Kroll et al. 2010; Olson which could potentially have negative consequences for et al. 2005). spotted owls (Glenn et al. 2010). Considering the potential The invasion of barred owls into the range of the effects of different climate change scenarios in models pre- northern spotted owl has been associated with a decreased dicting wildfire suitability (see chapter 4 in this report) may ability to detect and monitor spotted owls when barred owls help estimate potential changes in the fire regime within the are present (Dugger et al. 2009, Glenn et al. 2010, Kroll et owl’s range and thus potential threats to habitat. In addition, al. 2010, Olson et al. 2005). In addition, the detection of

87 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure 5-1—Annual proportion of northern spotted owl territories with barred owl detections on study areas in Washington, Oregon, and California. Source: adapted from Forsman et al. (2011).

88 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

the inclusion of forest type variables in the owl habitat mod- of nesting/roosting habitat for spotted owls, and the effects els can also be modified based on climate change scenarios of habitat loss cannot be decoupled from the additional and used to explore the effects of climate change on suitable stressor imposed by the barred owl range expansion (Dug- owl habitat. As noted by Glenn et al. (2010), however, in the ger et al. 2008). face of climate change and barred owl persistence, the best In particular, the relationship between spotted owl management strategy for conserving spotted owl popula- fitness and habitat characteristics may have become discon- tions is to maintain sufficient, high-quality, suitable habitat nected through interspecific competition with barred owls throughout the species’ range. in the landscape (Dugger et al. 2008). Our comparison of habitat suitability at spotted owl pair locations between Related Research and Research Needs 1994/96 and 2006/07 (see chapter 3 in this report) showed Current research efforts to further understand the com- an average decrease in habitat suitability value of 9.4 petitive interactions between barred and spotted owls are percent across the owl’s range, suggesting that the quality ongoing by D. Wiens, an Oregon State University Ph.D. of habitat at spotted owl pair locations has decreased over student. In addition, the U.S. Fish and Wildlife Service time. As loss of suitable nesting/roosting habitat since (FWS) is proceeding with a proposal to clarify interactions 1994/96 has been low (3.4 percent), it is unlikely this decline between the two species using an experimental approach in habitat quality of owl pair locations is the result of gener- as part of Recovery Action 29 in the Final Recovery Plan al habitat loss, so it is possible this change reflects competi- for the Northern Spotted Owl (USDI 2008). This recovery tion for space with barred owls. Barred owls will use a wide action calls for the FWS to “Design and implement large- variety of forested landscapes (Hamer et al. 2007, Singleton scale control experiments in key spotted owl areas to assess et al. 2010) and may be excluding spotted owls from the best the effects of barred owl removal on spotted owl site oc- spotted owl habitat in places where their densities are high cupancy, reproduction, and survival.” By removing barred (Dugger et al. 2008), but this hypothesis needs to be tested owls from spotted owl territories, researchers will be able to directly with barred owl removal experiments. It is possible document a clear cause-effect relationship between barred that competition with barred owls also might be the reason owl presence and spotted owl demography (Gutiérrez et al. we have had difficulty developing predictive models (see 2007). The information gained from this experiment may following discussion) that provide a clear understanding of aid in the management and conservation of spotted owls in the relationship between habitat characteristics and spotted the face of the continued threat posed by the invasion and owl demographics across the species’ range (Anthony et al. establishment of the barred owl in the Pacific Northwest. 1998, 2002a, 2002b; but see Dugger et al. 2005, Olson et al. This proposed research may elucidate new management 2004). actions or clarify management and conservation limitations The effectiveness monitoring plan for spotted owls regarding the negative interactions of these two species. recommended the development of predictive models linking The extent to which habitat management can affect survival, fecundity, and occupancy to observed vegetation interactions between barred and spotted owls is not clear, characteristics of owl habitat (Lint et al. 1999). The expecta- but barred owls are habitat generalists that can occupy a tion was that these predictive models could be validated and wide variety of forest conditions including late-successional proved to generate owl vital rate predictions with accept- forests (Herter and Hicks 2000, Pearson and Livezey 2003, able error. If so, then there would be a shift from intensive Singleton et al. 2010). In addition, their territories are only collection of mark-recapture data via annual field surveys 1/4 to 1/9 the size of spotted owl territories (Singleton et al. to the use of remotely sensed habitat data to monitor owl 2010), so the ratio of the number of barred to spotted owls populations on at least some of the eight study areas (Lint can be as high as 9 to 1 in some areas. This argues for the et al. 1999). The spotted owl monitoring program funded increased importance of high-quality, contiguous blocks a 5-year study to explore the development and feasibility

89 GENERAL TECHNICAL REPORT PNW-GTR-850

of predicting occupancy and demographic performance of owls is important for developing accurate estimates of site spotted owls using remotely sensed habitat data (Anthony occupancy (MacKenzie et al. 2006, Olson et al. 2005). In et al. 1998), and most of that work has been completed addition, understanding the mechanisms or processes that (Anthony et al. 2002a, 2002b; Dugger et al. 2005, 2006, drive site occupancy, like the factors that affect the prob- 2008; Olson et al. 2004, 2005). ability that an occupied site becomes unoccupied (i.e., local Unfortunately, this component of the effectiveness extinction rate) or the probability that an unoccupied site monitoring program has produced mixed results, with only becomes occupied (i.e., local colonization rate) are proving a few strong relationships between habitat characteristics vital to understanding the impact of barred owls and habitat and survival and fecundity noted for some of the demo- characteristics on spotted owl persistence (Olson et al. 2005; graphic study areas (Dugger et al. 2005, Franklin et al. Kroll et al. 2010; Dugger et al., in press). Based on these 2000, Olson et al. 2004). As noted by Lint (2005) in the 10- models, strong relationships between the amount of old- year report, results at that point did not warrant moving on forest habitat at the core scale (410-ac circle around nest tree to phase II monitoring, where models would be substituted or activity center) and extinction rates were observed for for mark-recapture studies. However, although simple, uni- the South Cascades study area; spotted owl territories with versal models linking habitat characteristics to survival and small amounts of old forest near the site center experienced fecundity of owls are likely not possible, these efforts have higher extinction rates of owl pairs (Dugger et al. 2008). In provided more insight into the effects of climate and habitat addition, increased fragmentation of old forest at the home characteristics on owl demography. Some general findings range scale (3,700-ac circle around nest site or activity include the strong positive effect of late-successional forest center) decreased colonization rates by owl pairs, and both at the core of an owl’s territory (around the nest site or ac- occupancy parameters were affected by barred owl presence tivity center) on survival and fecundity (Dugger et al. 2005, as well (Dugger et al. 2008). Franklin et al. 2000, Olson et al. 2004). In addition, at least It is unclear why we observed stronger associations on some study areas in the southern portion of the owls’ between habitat characteristics and occupancy parameters range, some component of edge habitat may be important, as compared to habitat characteristics and survival or probably as a source of prey (Franklin et al. 2000, Olson et fecundity (Anthony et al. 2002a, 2002b; Dugger et al. 2005, al. 2004). 2006, 2008; Olson et al. 2004, 2005), but it is possible that Since the 10-year report, models for two study areas occupancy reflects the first level of selection by a species, have been developed linking occupancy dynamics of and this is where the strongest selections for habitat are spotted owls to habitat characteristics (Dugger et al. being made. In other words, an area of habitat selected 2008; Sovern, 2010). The effect of barred owls and habitat for defense and maintenance of a territory by an owl also characteristics on extinction and colonization rates can meets some minimum standard of suitability for survival be modeled using multiseason, single-species occupancy and reproduction; thus, habitat quality most strongly affects models (MacKenzie et al. 2006) or even multiple-species territory selection, but other factors (climate, age/experience models within seasons (Bailey et al. 2009, MacKenzie et of individuals, individual variation) are more important for al. 2006). Estimates of annual site occupancy can also be explaining the variation in survival and fecundity. derived from these models, which rely on a mark-recapture Recent advances in the development of remotely sensed framework with a “site” or owl territory as the sample unit vegetation (Ohmann and Gregory 2002) and change-detec- and presence/absence data across multiple visits within and tion data (Kennedy et al. 2007) may provide an opportunity between years to allow the separation of occupancy dynam- to investigate habitat relationships across the range of the ics and detection probabilities (MacKenzie et al. 2006). species in conjunction with barred owl influences. Previous Accounting for variations in detection rates of spotted efforts included a range of map products based on a single

90 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

point in time, or limited temporally and of varying qual- wildfires might affect efforts to conserve the owl. Hotter, ity (Glenn and Ripple 2004), precluding a meta-analysis drier climates associated with climate change are believed using data from all the study areas. The development of to be at least partially responsible for this increase in large- this new vegetation layer will now allow us to search for wildfire frequency (Westerling et al. 2006), and there is and quantify consistent relationships between habitat also evidence that the amount of high-severity wildfire has characteristics and owl demography, particularly occupancy increased (Miller et al. 2009, Schwind 2008; but see Hanson across the entire range of the species within a meta-analysis et al. 2009), in some cases, as the result of accumulated framework. In addition, the change-detection data provide fuels and higher stand densities (Sensenig 2002). an annual time sequence of vegetation changes that can now The relationship between wildfire and owl demogra- be linked to annual demographic data. This kind of analysis phy is not well understood, but likely includes a complex based on data from eight effectiveness monitoring areas and interaction of fire frequency and severity (Bond et al. conducted in a workshop format as a meta-analysis follow- 2009, Clark 2007). Owls use forest stands that have burned ing previous efforts for survival and fecundity (Anthony et understories or partially removed overstories, but they tend al. 2006, Burnham et al. 1996, Forsman et al. 2011, Franklin to avoid areas of complete stand replacement for nesting and et al. 1999) should be a priority for future research. roosting (Clark 2007), although use of high-severity burn A better understanding of the population dynamics of areas for foraging has been documented for the California many of the important spotted owl prey species across the spotted owl (Strix occidentalis occidentalis) (Bond et al. range of the owl will likely be essential to understanding 2009). This species has likely evolved the ability to adapt patterns and variation in spotted owl fecundity (Courtney et and utilize forests that have been subjected to light to al. 2004, Forsman et al. 2011). New monitoring and research moderate fire severity, particularly in the fire-prone portions programs should be initiated to investigate prey cycles and of its range (chapter 4, this report), but again short-term their relationship to spotted owl demographics while incor- vs. long-term effects on demography and dispersal are porating the potential competitive effects of barred owls. unknown. This remains a large gap in our understanding of spotted Although wildfire has long been a natural agent of owl ecology, and our lack of baseline information increases disturbance, owls evolved with it in historically forested the difficulty we face trying to manage spotted owl popula- landscapes that could accommodate the habitat changes tions in conjunction with the barred owl, most likely a direct caused by it. Today, much of the spotted owl habitat that competitor for food resources. remains has been “squeezed” into federally managed Another area of much needed research includes the lands, covers a much smaller portion of the owl’s historical effect of fire on owls and their prey, and how fuel reduc- range, is highly fragmented (Davis and Lint 2005), and tion treatments proposed to reduce wildfire risk affect owl may no longer be able to accommodate large wildfires demography. Fire suppression over the last century has without incurring adverse consequences to the owl. To reduced wildfire’s presence in its “natural habitats” (Agee lessen the chances of adverse impacts from occurring, the 1993, Atzet and Martin 1992, Sensenig 2002), and although Final Recovery Plan for the Northern Spotted Owl (USDI wildfire risk has not increased dramatically in the moister/ 2008) advocated landscape-level treatments to reduce the cooler forests, this suppression is believed to have increased risk of large-scale habitat loss to high-severity wildfire for the risk for severe wildfires in the fire-prone, or drier/warm- Eastern Cascades and Klamath provinces of the owl’s range er forests. The increased frequency of large wildfires since (USDI 2008). However, it is currently unclear what short- or the mid-1980s in the Western United States (Westerling et long-term effects these forest thinning and fuel reduction al. 2006, Schwind 2008) and within the owl’s range (see treatments will have on northern spotted owl populations. chapter 4 in this report) have created concern about how

91 GENERAL TECHNICAL REPORT PNW-GTR-850

A case study on a single owl territory in second-growth has been responsible for documentation of the barred owl forests in the northern Oregon Coast Range suggests com- expansion southward into the spotted owl’s range and the mercial thinning may cause northern spotted owls to alter negative effects of this invasion on spotted owls (Dugger their habitat use and increase the size of their home ranges, et al. 2008, 2009; Olson et al. 2005). Recovery goals and particularly during the nonbreeding season (Meiman et al. actions associated with the Final Recovery Plan for the 2003). This one case study suggested that thinning opera- Northern Spotted Owl (USDI 2008) and proposed revisions tions within core-use areas may be detrimental for northern have been informed directly by or are reliant on the demog- spotted owls, at least in the short term. But it is not known raphy data collected on the eight effectiveness monitoring whether thinning produces long-lasting adverse impacts study areas, as well as the remotely sensed data developed or long-term benefits associated with owl vital rates. No for habitat monitoring. Data from this long-term monitoring other published literature is available on thinning and the program have also aided researchers in the development of effects of fuel reductions on habitat use and demography new analytical approaches for answering complex demog- of threatened spotted owls. Understanding the relationship raphy questions (Bailey et al. 2009, MacKenzie et al. 2006). between wildfire and owl demography and the effect of These examples illustrate how the value of the spotted owl both commercial and noncommercial thinning activities effectiveness monitoring program reaches far beyond the (to reduce fire fuel loads) on owl vital rates should be high original objectives and is truly vital to management and research priorities. conservation of this species.

Summary References As we have summarized above, there are several large gaps Agee, J.K. 1993. Fire ecology of Pacific Northwest forests. in our understanding of spotted owl ecology, particularly in Washington, DC: Island Press. 493 p. relation to cycles of prey distribution and abundance, distur- Anthony, R.G.; Forsman, E.D.; Franklin, A.B.; bance by fire, and forest management activities associated Anderson, D.R.; Burnham, K.P.; White, G.C.; with developing future habitat or reducing fire risk. Emerg- Schwarz, C.J.; Nichols, J.D.; Hines, J.E.; Olson, ing issues, primarily the competitive interactions with the G.S.; Ackers, S.H.; Andrews, L.W.; Biswell, B.L.; barred owl, are also of very high concern, particularly as the Carlson, P.C.; Diller, L.V.; Dugger, K.M.; Fehring, negative effect of this invasive species may be in addition K.E.; Fleming, T.L.; Gearhardt, R.P.; Gremel, to, or somewhat independent of, maintenance of high-qual- S.A.; Gutiérrez, R.J.; Happe, P.J.; Herter, D.R.; ity spotted owl habitat. The information we have on these Higley, J.M.; Horn, R.B.; Irwin, L.L.; Loschl, P.J.; issues is dependent on continued research and, in particular, Reid, J.A.; Sovern, S.G. 2006. Status and trends in the continued long-term monitoring of owl vital rates demography of northern spotted owls, 1985–2003. throughout this species’ range. The effectiveness monitor- Wildlife Monographs No. 163. Washington, DC: The ing program for spotted owls was designed to monitor the Wildlife Society. 48 p. long-term results of the Plan and its effect on owl populations (Lint et al. 1999). This monitoring program has done Anthony, R.G.; Forsman, E.D.; Ripple, W.R. 1998. much more, however, as the unique, large-scale demography Research proposal: predicting abundance and demo- data set resulting from this program has not only allowed graphic performance of northern spotted owls from resource managers to document the effects of management vegetative characteristics. 20 p. Unpublished research activities, but has also contributed valuable information proposal. On file with: Oregon Cooperative Fish and regarding basic owl ecology and the factors that affect vital Wildlife Research Unit, Department of Fisheries and rates. In large part, the effectiveness monitoring program Wildlife, 104 Nash Hall, Oregon State University, Corvallis, OR 97331.

92 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Anthony, R.G.; Forsman, E.D.; Ripple, W.R.; Glenn, Carroll, C. 2010. Role of climatic niche models in focal- E. 2002a. Predicting abundance and demographic species-based conservation planning: assessing potential performance of northern spotted owls from vegetative effects of climate change on northern spotted owls in the characteristics. Report on results of Western Oregon Pacific Northwest, USA. Biological Conservation. 143(6): Cascades–H.J. Andrews Experimental Forest. 57 p. 1432–1437. Unpublished progress report. On file with: Oregon Carroll, C.; Dunk, J.R.; Moilanen, A. 2009. Optimizing Cooperative Fish and Wildlife Research Unit, Depart- resiliency of reserve networks to climate change: ment of Fisheries and Wildlife, 104 Nash Hall, Oregon multispecies conservation planning in the Pacific State University, Corvallis, OR 97331. Northwest, USA. Global Change Biology. 16: 891–904. Anthony, R.G.; Wagner, F.; Dugger, K.M.; Olson, Clark, D.A. 2007. Demography and habitat selection G. 2002b. Identification and evaluation of northern of northern spotted owls in post-fire landscapes of spotted owl habitat in managed forests of southwestern southwestern Oregon. Corvallis, OR: Oregon State Oregon and the development of silvicultural systems University. 218 p. M.S. thesis. for managing such habitat. Report on step 2: analysis of habitat characteristics and owl demography on three Courtney, S.P.; Blakesley, J.A.; Bigley, R.E.; Cody, density study areas. 77 p. Unpublished report. On file M.L.; Dumbacher, J.P.; Fleischer, R.C.; Franklin, with: Oregon Cooperative Fish and Wildlife Research A.B.; Franklin, J.F.; Gutiérrez, R.J.; Marzluff, J.M.; Unit, Department of Fisheries and Wildlife, 104 Nash Sztukowski, L. 2004. Scientific evaluation of the status Hall, Oregon State University, Corvallis, OR 97331. of the northern spotted owl. Portland, OR: Sustainable Ecosystem Institute. 348 p. + appendixes. Atzet, T.A.; Martin, R. 1992. Natural disturbance regimes in the Klamath province. In: Kerner, H.M., Dugger, K.M.; Anthony, R.G.; Andrews, L.S. [In press]. ed. Proceedings of the symposium on biodiversity of Transient dynamics of invasive competition: barred owls, northwestern California. Report 29. Berkeley, CA: spotted owls, and the demons of competition present. University of California Wildland Resources Center: Ecological Applications. 40–48. Dugger, K.M.; Anthony, R.G.; Andrews, L.S.; Wagner, Bailey, L.L.; Reid, J.A.; Forsman, E.D.; Nichols, J.D. F. 2006. Modeling apparent survival and reproductive 2009. Modeling co-occurrence of northern spotted success of northern spotted owls relative to forest habitat and barred owls: accounting for detection probability in the southern Cascades of Oregon. 45 p. Unpublished differences. Biological Conservation. 142: 2983–2989. report. On file with: Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Bond, M.L.; Lee, D.E.; Siegel, R.B.; Ward, J.P., Jr. 2009. Wildlife, 104 Nash Hall, Oregon State University, Habitat use and selection by California spotted owls in Corvallis, OR 97331. a postfire landscape. Journal of Wildlife Management. 73(7): 1116–1124. Dugger, K.M.; Anthony, R.G.; Forsman, E.D. 2009. Estimating northern spotted owl detection probabilities: Burnham, K.P.; Anderson, D.R.; White, G.C. 1996. updating the USFWS northern spotted owl survey Meta-analysis of vital rates of the northern spotted owl. protocol. 55 p. Unpublished report. On file with: Studies in Avian Biology. 17: 92–101. Oregon Cooperative Fish and Wildlife Research Unit, Department of Fisheries and Wildlife, 104 Nash Hall, Oregon State University, Corvallis, OR 97331.

93 GENERAL TECHNICAL REPORT PNW-GTR-850

Dugger, K.M.; Anthony, R.G.; Lawrence, L.S. 2008. Site Franklin, A.B.; Burnham, K.P.; White, G.C.; Anthony, occupancy dynamics of northern spotted owls relative to R.G.; Forsman, E.D.; Schwarz, C.; Nichols, J.D.; barred owl occurrence and forest habitat characteristics Hines, J. 1999. Range-wide status and trends in northern in the southern Cascades of Oregon. 40 p. Unpublished spotted owl populations. 71 p. Unpublished report. report. On file with: Oregon Cooperative Fish and On file with: Colorado Cooperative Fish and Wildlife Wildlife Research Unit, Department of Fisheries and Research Unit, Colorado State University, 201 Wagar Wildlife, 104 Nash Hall, Oregon State University, Building, 1484 Campus Delivery, Fort Collins, Colorado Corvallis, OR 97331. 80523. http://www.reo.gov/monitoring/reports/nso/ Dugger, K.M.; Wagner, F.; Anthony, R.G.; Olson, G.S. N110202_3_1999-REPORT-FINAL-DRAFTREV.pdf. (4 2005. The relationship between habitat characteristics January 2011). and demographic performance of northern spotted owls Funk, W.C.; Forsman, E.D.; Johnson, M.; Mullins, in southern Oregon. Condor. 107: 863–878. T.D.; Haig, S.M. 2010. Evidence for recent population Fitzgerald, S.D.; Patterson, J.S.; Kiupel, M.; Simmons, bottlenecks in northern spotted owls (Strix occidentalis H.A.; Grimes, S.D.; Sarver, C.F.; Fulton, R.M.; caurina). Conservation Genetics. 11: 1013–1021. Steficek, B.M.; Cooley, T.M.; Massey, J.P.; Sikarskie, Glenn, E.M.; Anthony, R.G.; Forsman, E.D. 2010. J.G. 2003. Clinical and pathologic features of West Nile Population trends in northern spotted owls: associations virus infection in native North American owls (Family with climate in the Pacific Northwest. Biological Strigidae). Avian Diseases. 47: 602–610. Conservation. 143: 2543–2552. Forsman, E.D.; Anthony, R.G.; Dugger, K.M.; Glenn, Glenn, E.M.; Ripple, W.J. 2004. On using digital maps to E.M.; Franklin, A.B.; White, G.C.; Schwarz, C.J.; assess wildlife habitat. The Wildlife Society Bulletin. 32: Burnham, K.P.; Anderson, D.R.; Nichols, J.D.; Hines, 852–860. J.E.; Lint, J.B.; Davis, R.J.; Ackers, S.H.; Andrews, Gutiérrez, R.; Cody, M.; Courtney, S.; Franklin, A.B. L.S.; Biswell, B.L.; Carlson, P.C.; Diller, L.V.; 2007. The invasion of barred owls and its potential Gremel, S.A.; Herter, D.R.; Higley, J.M.; Horn, R.B.; effect on the spotted owl: a conservation conundrum. Reid, J.A.; Rockweit, J.; Schaberl, J.; Snetsinger, T.J.; Biological Invasions. 9: 181–196. Sovern, S.G. 2011. Population demography of northern spotted owls. Studies in Avian Biology 40. Berkeley, CA: Hamer, T.E.; Forsman, E.D.; Glenn, E.M. 2007. Home University of California Press. 106 p. range attributes and habitat selection of barred owls and spotted owls in an area of sympatry. Condor. 109: Franklin, A.B. 2010. Personal communication. Project 750–768. leader, Ecology of Emerging Viral and Bacterial Diseases in Wildlife Project. USDA/APHIS/WS National Hanson, C.T.; Odion, D.C.; Dellasala, D.A.; Baker, B.L. Wildlife Research Center, 4101 Laporte Ave., Fort 2009. Overestimation of fire risk in the northern spotted Collins, CO 80521–2154. owl recovery plan. Conservation Biology. 23: 1314–1319. Franklin, A.B.; Anderson, D.R.; Gutiérrez, R.J.; Herter, D.R.; Hicks, L.L. 2000. Barred owl and spotted Burnham, K.P. 2000. Climate, habitat quality, owl populations and habitat in the central Cascade Range and fitness in northern spotted owl populations in of Washington. Journal of Raptor Research. 34: 279–286. northwestern California. Ecological Monographs. 70(4): Kennedy, R.E.; Cohen, W.B.; Schroeder, T.A. 2007. 539–590. Trajectory-based change-detection for automated characterization of forest disturbance dynamics. Remote Sensing of Environment. 110: 370–386.

94 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Kroll, A.J.; Fleming, T.L.; Irwin, L.L. 2010. Site Olson, G.S.; Anthony, R.G.; Forsman, E.D.; Ackers, occupancy dynamics of northern spotted owls in the S.H.; Loschl, P.J.; Reid, J.A.; Dugger, K.M.; Glenn, eastern Cascades, Washington, USA, 1990–2003. E.M.; Ripple, W.J. 2005. Modeling of site occupancy Journal of Wildlife Management. 74: 1264–1274. dynamics for northern spotted owls, with emphasis on the effects of barred owls. Journal of Wildlife Lint, J.; Noon, B.; Anthony, R.; Forsman, E.; Raphael, Management. 69: 918–932. M.; Collopy, M.; Starkey, E. 1999. Northern spotted owl effectiveness monitoring plan for the Northwest Olson, G.S.; Glenn, E.M.; Anthony, R.G.; Forsman, Forest Plan. Gen. Tech. Rep. PNW-GTR-440. Portland, E.D.; Reid, J.A.; Loschl, P.J.; Ripple, W.J. 2004. OR: U.S. Department of Agriculture, Forest Service, Modeling demographic performance of northern spotted Pacific Northwest Research Station. 43 p. owls relative to forest habitat in Oregon. Journal of Wildlife Management. 68: 1039–1053. Lint, J.B., tech. coord. 2005. Northwest Forest Plan—the first 10 years (1994–2003): status and trends of northern Pearson, R.R.; Livezey, K.B. 2003. Distribution, numbers, spotted owl populations and habitat. Gen. Tech. Rep. and site characteristics of spotted owls and barred owls PNW-GTR-648. Portland, OR: U.S. Department of in the Cascade Mountains of Washington. Journal of Agriculture, Forest Service, Pacific Northwest Research Raptor Research. 37: 265–276. Station. 176 p. Schwind, B., comp. 2008. Monitoring trends in burn Livezey, K.B. 2009. Range expansion of barred owls. Part severity: report on the Pacific Northwest and Pacific I: Chronology and distribution. American Midland Southwest fires (1984 to 2005). Unpublished report. Naturalist. 161: 49–56. On file with: USDA Forest Service, Remote Sensing Applications Center (RSAC), U.S. Geological Survey, MacKenzie, D.I.; Nichols, J.D.; Royle, J.A.; Pollock, Center for Earth Resources Observations and Science K.H.; Bailey, L.L.; Hines, J.E. 2006. Occupancy (EROS). http://mtbs.gov/reports/MTBS_pnw-psw_final. estimation and modeling: inferring patterns and pdf. (4 January 2011). dynamics of species occurrence. New York, NY: Academic Press. 344 p. Sensenig, T.S. 2002. Development, fire history and current and past growth, of old-growth and young-growth forest Meiman, S.; Anthony, R.; Glenn, E.; Bayless, T.; stands in the Cascade, Siskyou and mid-Coast Mountains Ellingson, A.; Hansen, M.C.; Smith, C. 2003. Effects of southwestern Oregon. Corvallis, OR: Oregon State of commercial thinning on home-range and habitat-use University. 180 p. Ph.D. dissertation. patterns of a male northern spotted owl: a case study. Wildlife Society Bulletin. 31(4): 1254–1262. Singleton, P.H.; Lemkuhl, J.F.; Gaines, W.L.; Graham, S.A. 2010. Barred owl space use and habitat selection in Miller, J.D.; Safford, H.D.; Crimmins, M.A.; Thode, the eastern Cascades, Washington. Journal of Wildlife A.E. 2009. Quantitative evidence for increasing forest Management. 74: 285–294. fire severity in the Sierra Nevada and southern Cascade Mountains, California and Nevada, USA. Ecosystems. Sovern, S.G. 2010. Personal communication. Wildlife 12: 16–32. biologist, Cle Elum demographic survey crew leader, USDA Forest Service, Pacific Northwest Research Ohmann, J.L.; Gregory, M.J. 2002. Predictive mapping Station, 802 West Second Street, Cle Elum, WA 98922. of forest composition and structure with direct gradient analysis and nearest-neighbor imputation in coastal Oregon, U.S.A. Canadian Journal of Forest Research. 32: 725–741.

95 GENERAL TECHNICAL REPORT PNW-GTR-850

Spies, T.A.; Giesen, T.W.; Swanson, F.J.; Franklin, Andrews, and Peter Carslon, who, in addition to making the J.F.; Lach, D.; Johnson, K.N. 2010. Climate change field work happen, took the time to review and comment on adaptation strategies for federal forests of the Pacific the habitat models that we generated from their data. We Northwest, USA: ecological, policy, and socio-economic thank them. We also thank Diana Stralberg, Renée Cormier, perspectives. Landscape Ecology. 25: 1185–1199. and Dave Press for sharing their northern spotted owl data U.S. Department of Agriculture [USDA]. 2009. Forest from the southernmost tip of its range. Service global change research strategy: 2009–2019. For the habitat monitoring, we thank Janet Ohmann, Washington, DC: U.S. Department of Agriculture, Matthew Gregory, and Heather Roberts, also known as the Forest Service, Research and Development. 20 p. Landscape, Ecology, Modeling, Mapping, and Analysis http://www.fs.fed.us/research/climate/strategy.shtml. team for providing us with our first consistent rangewide (17 December 2010). vegetation data sets. This could not have been done without the collaborative efforts of Robert Kennedy, Warren Cohen, U.S. Department of the Interior [USDI]. 2008. Final and Zhiqiang Yang from the Laboratory for Applications of recovery plan for the northern spotted owl (Strix Remote Sensing in Ecology research laboratory, which also occidentalis caurina). 142 p. On file with: U.S. Depart- produced the change-detection data for our analysis. This ment of the Interior, Fish and Wildlife Service, 2600 SE group was incredibly helpful to our modeling efforts and 98th Ave., Suite 100, Portland, OR 97266. always available to answer questions and offer advice. We Westerling, A.L.; Hidalgo, H.G.; Cayan, D.R.; Swetnam, are grateful for the collaborative support we received from T.W. 2006. Warming and earlier spring increase Western the Marbled Murrelet Monitoring Team, our close partners U.S. forest fire activity. Science. 313: 940–943. in habitat monitoring, particularly Martin Raphael, Gary Falxa, Beth Galleher, and Rich Young. Thanks to Carol Acknowledgments Apple for her advice and assistance in the inventory plot This report is the result of a large group effort focused analysis. We are also grateful for the data and advice pro- on monitoring the effectiveness of the Northwest Forest vided by Alexandre Hirzel and Stephen Phillips (developers Plan (the Plan). It would not have been possible without of the BioMapper and MaxEnt habitat modeling software, the continued hard work of many dedicated biologists who respectively), who patiently answered our many questions as annually collect the field data and help to generate the we experimented with different habitat modeling methods natural resource information that we periodically analyze to during the early phases of this report. produce these monitoring reports on the status and trends The chapter on wildfire suitability was the result of a of northern spotted owls and how they are faring under long-term collaboration with several people from several the Plan. This unique set of field data is the cornerstone of agencies, including Joe Lint (retired), Jim Thrailkill, Lynn our population monitoring and would not exist without the Gemlo, Cindy Donegan, Scott Center, Elaine Rybak, and vision and leadership of the principal investigators of the Josh Chapman. We thank them for the many discussions, demography study areas—Robert Anthony, USGS (retired); ideas, and assistance that eventually led to us being able Eric Forsman, USDA Forest Service; Alan Franklin, to portray large wildfires in this unique ecological and Colorado State University: Rocky Gutierrez, University of geographical perspective. Minnesota; and their many research associates. We thank This report benefited from the helpful comments of them for their tremendous contribution to this effort. The several reviewers who read various drafts or chapters of federal demographic survey area crew leaders (from north it, including Tom Spies, Robert Anthony, Elizabeth Glenn, to south) are Brian Biswell, Scott Gremel, Stan Sovern, Tom Warren Cohen, Carl Skinner, and Ayn Shlisky. We thank Snetsinger, Steve Ackers, Janice Reid, Robert Horn, Steve Ashley Steel for the statistical review and Jim Baldwin for

96 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

his early comments and assistance in developing a spread- sheet for calculating the Continuous Boyce Index. A special thanks to Shawne Mohoric (retired) and all those in- volved in the Interagency Regional Monitoring Program— especially Melinda Moeur (retired) for her efforts to continually advance and improve the vegetation data that several modules use for monitoring. We greatly appreciate the budget and personnel support from the federal partner- ship that made this all happen, as well as the Interagency Senior Managers Group and the Regional Interagency Executive Committee for their continuing commitment and support for the Northwest Forest Plan effectiveness monitoring program. And last, but certainly not least, we thank Joe Lint for laying down a solid spotted owl monitoring foundation upon which we hope to continue to build.

Metric Equivalents When you know: Multiply by: To find:

Inches (in) 25.4 Millimeters (mm) Inches (in) 2.54 Centimeters (cm) Feet (ft) .305 Meters (m) Miles (mi) 1.609 Kilometers (km) Square miles (mi2) 2.59 Square kilometers (km2) Acres (ac) .405 Hectares (ha) Degrees Fahrenheit (°F) (F-32)/1.8 Degrees Celsius (°C)

97 GENERAL TECHNICAL REPORT PNW-GTR-850 Peter Carlson Peter

Barred owl.

98 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix A: Environmental Variables Used for Habitat Suitability Modeling

Table A-1—Environmental variables that were used for habitat modeling Variable Description Units Diameter A measure of the structural diversity of a forest stand based on tree densities Index diversity in different diameter at breast height (d.b.h.) classes. Calculation procedures index are described in appendix 1 of McComb et al. (2002)a Canopy cover Percentage of conifer cover in the canopy as calculated using methods in the Percentage of all conifers Forest Vegetation Simulator Stand height Average height of dominant and codominant trees Meters Mean conifer Basal area weighted mean diameter of all live conifers Centimeters diameter Density of Estimated tree density for all live conifers ≥ 30 in d.b.h. Trees/ha large conifers Stand age Average stand age based on field-recorded ages of dominant and codominant (no remnants) tree species, and excluding remnant trees Years Subalpine Stand component of Pacific silver firAbies ( amabilis Dougl. ex Forbes), Percentage forest subalpine fir Abies( lasiocarpa (Hook.) Nutt.), noble fir Abies( procera of total Rehd.), Shasta red fir (Abies shastensis (Lemmon) Lemmon), Alaska cedar basal area (Chamaecyparis nootkatensis (D. Don) Spach), Engelmann spruce (Picea engelmannii Parry ex Engelm.), whitebark pine (Pinus albicaulis Engelm.), and mountain hemlock (Tsuga mertensiana (Bong.) Carr.) Pine forest Stand component of lodgepole pine (Pinus contorta Dougl. ex Loud.), Percentage Jeffrey pine (Pinus jeffreyi Grev. & Balf.), Bishop pine (Pinus muricata of total D. Don), and ponderosa pine (Pinus ponderosa Dougl. ex Laws.) basal area Oak woodlands Stand component of blue oak (Quercus douglasii Hook. & Arn.), Percentage Oregon white oak (Quercus garryana Dougl. ex Hook.), and California of total black oak (Quercus kelloggii Newb.) basal area Evergreen Stand component of Pacific madrone Arbutus( menziesii Pursh), tanoak Percentage hardwoods (Lithocarpus densiflorus Rehd.), California live oak (Quercus agrifolia Née), of total canyon live oak (Quercus chrysolepis Liebm.), and California laurel basal area (Umbellularia californica (Hook. & Arn.) Nutt.) Redwood forest Stand component of redwood (Sequoia sempervirens D. Don) Endl.) Percentage of total basal area a McComb, W.C.; McGrath, M.T.; Spies, T.A.; Vesely, D. 2002. Models for mapping potential habitat at landscape scales: an example using northern spotted owls. Forest Science. 48(2): 203–216.

99 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure A-1–Stand structure and age habitat variable correlation matrix with averaged accuracy plot Pearson correlations (SD = standard deviation, d.b.h. = diameter at breast height). These six environmental variables were used in all modeling regions.

100 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

kappa 0.59 0.48 0.40 0.25 0.33 0.37 0.48 0.33 0.23 0.28 0.51 0.55 0.41 0.42 0.41 0.61 0.31 0.30 0.34 0.59 Average

n/a n/a n/a n/a n/a n/a n/a 0.55 0.30 0.21 0.14 0.28 0.48 0.41 0.34 0.52 0.29 0.31 0.22 0.59 Coast (MR 226) California

n/a n/a n/a n/a n/a n/a n/a n/a 0.58 0.29 0.47 0.26 0.28 0.28 0.34 0.68 0.35 0.38 0.43 0.35 (MR 225) Klamaths California Oregon and Oregon –

n/a n/a n/a n/a n/a n/a 0.59 0.39 0.52 0.19 0.22 0.34 0.62 0.57 0.27 0.58 0.52 0.53 0.45 0.17 Cascades (MR 224) California Oregon and Oregon

Kappa coefficients

n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0.29 0.27 0.49 0.72 0.46 0.43 Coast Range Oregon (MR 223)

n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0.66 0.58 0.29 0.29 0.38 0.46 0.53 0.26 0.62 0.56 Eastern Cascades (MR 222) Washington

n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a n/a 0.53 0.48 0.32 0.28 0.38 0.32 0.50 – Cascades

(MR 221) Coast and Washington

Pacific silver fir Subalpine fir Noble fir/Shasta red fir Alaska cedar Engelmann spruce Whitebark pine Mountain hemlock Lodgepole pine pine Jeffrey Bishop pine Ponderosa pine Blue oak Oregon white oak California black oak Pacific madrone Tanoak California live oak Canyon live oak California laurel Redwood

Common name

Abies amabilis Abies lasiocarpa Abies procera/shastensis Chamaecyparis nootkatensis Picea engelmannii Pinus albicaulis mertensiana Tsuga Pinus contorta Pinus jeffreyi Pinus muricata Pinus ponderosa douglasii Quercus garryana Quercus kelloggii Quercus Arbutus menziesii Lithocarpus densiflorus agrifolia Quercus chrysolepis Quercus Umbellularia californica Sequoia sempervirens Scientific name

GNN ABAM ABLA ABPRSH CHNO PIEN PIAL TSME PICO PIJE PIMU PIPO QUDO QUGA4 QUKE ARME LIDE3 QUAG QUCH2 UMCA SESE3

DOM SPP

forest forest woodlands hardwoods Table A-2—StandTable species composition variable groupings, with local scale accuracy assessments, used in applicable modeling regions (MR) (GNN = gradient dominantDOM SPP nearest species neighbor, models) Species grouping Subalpine Pine Oak Evergreen Redwood forest Note: n/a = not applicable.

101 GENERAL TECHNICAL REPORT PNW-GTR-850 Stan Sovern

102 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix B: Nearest Neighbor Distance Analysis of Demographic Study Area Data

An analysis of nearest neighbor distances (Clark and Evans 1954) was conducted on several demographic study area owl pair location (fig. B–1) data sets from 1994 through 1997 to correspond with the baseline satellite imagery. The purpose of this analysis was to determine biologically relevant distances for use as minimum distance parameters in the random sampling of owl pair locations from the 10-year report owl presence data set (Davis and Lint 2005). The purpose of this sampling was to provide additional habitat model training data points, outside of demographic study areas, for the habitat modeling described in chapter 3. Only one location was used to represent each owl territory center. To minimize erroneous results, we only used owl locations from the 50-percent harmonic mean core (Dixon and Chapman 1980) of each study area’s data set. This removed outlier locations that would introduce errors in the analysis, especially for study areas that have disjunct areas or survey areas that are separated by several miles. The analysis was conducted in ArcView Spatial Analyst using the Animal Movement extension (v2.0) by Hooge and Eichenlaub (2000). Results show a decreasing trend in distance between owl pair territories from north to south (fig. B-2). The greatest mean nearest neighbor distance occurs in the Wash- ington Eastern Cascades (4.5 km), and the shortest mean distance occurs within the California Coast (1.4 km). The longer distances in the northern portions of the range may relate to more limited prey resources. Likewise the shorter distances in the southern portion of the range may be due to increased prey base diversity and abundance associated with the presence of mast-producing evergreen hardwoods that occur in the coniferous forests of that region.

Figure B-1—Owl territories, 1994 through 1997.

103 GENERAL TECHNICAL REPORT PNW-GTR-850

Table B-1–Summary statistics from the nearest neighbor analysis for each habitat modeling region Mean Median Modeling region Data sets used na distance distance StDev – – – – – Kilometers – – – – – Washington Eastern Cle Elum and T. Fleming study 26 4.5 3.9 2.4 Cascades areas Washington Coast Olympic and Rainier study areas 53 3.6 3.1 1.7 and Cascades Oregon and California H.J. Andrews and Southern 57 3.1 2.9 1.5 Cascades Cascades study areas Oregon Coast Range Oregon Coast Ranges and Tyee 79 2.7 2.5 1.2 study areas Oregon and California Klamath, Northwest California, 70 2.4 2.1 1.1 Klamaths and Hoopa study areas California Coast Green Diamond Resources 77 1.4 1.3 0.6 and Marin study areas a Only locations from the 50-percent harmonic core of the study area data set were used. StDev = standard deviation.

Figure B-2—Results of the nearest neighbor distance analysis showing mean distances between northern spotted owl territory centers with 95-percent confidence intervals for each habitat modeling region.

104 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

References Dixon, K.R.; Chapman, J.A. 1980. Harmonic Clark P.; Evans, F.C. 1954. Distance to the nearest mean measure of animal activity areas. Ecology. neighbor as a measure of spatial relationship in 61: 1040–1044. populations. Ecology. 35: 445–453. Hooge, P.N.; Eichenlaub, B. 2000. Animal movement Davis, R.; Lint, J. 2005. Habitat status and trend. In: extension to ArcView, 2.0. Anchorage, AK: Alaska Lint, J., tech. coord. Northwest Forest Plan—the first Science Center—Biological Science Office, U.S. 10 years (1994–2003): status and trends of northern Geological Survey. spotted owl populations and habitat. Gen. Tech. Rep. PNW-GTR-648. Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station: 21–82. Janice Reid

105 GENERAL TECHNICAL REPORT PNW-GTR-850 Patrick Kolar

106 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix C: Habitat Suitability Modeling Replicate Data

Pine cover forest forest forest forest Conifer gain the gain removed Lowered Subalpine Subalpine Subalpine hardwood Evergreen most when

age index cover index index Stand Conifer by itself diversity diversity diversity diversity Diameter Diameter Diameter Highest gain

n/a n/a n/a n/a 6.2 n/a

forest Redwood

n/a 1.9 2.6 10.9 10.2 n/a

hardwood Evergreen

n/a 0.9 1.1 4.4 13.0 1.3 Oak woodland

n/a n/a 5.5 18.3 5.7 4.9 Species composition variables Pine forest

–

10.8 n/a 6.3 2.4 n/a 31.8 forest Subalpine

20.2 11.2 4.7 5.1 2.3 2.7 age Stand

Percentage of contribution Percentage 3.2 4.6 3.8 8.9 6.3 7.4 Stand

height

7.8 5.4 21.7 14.1 2.6 6.0

index diversity Diameter

35.2 58.1 33.4 15.0 3.0 11.0 Large conifer density

1.7 4.6 5.2 1.8 7.1 0.5 d.b.h. Mean conifer Stand structure and age variables Stand structure

21.0 13.3 15.6 19.1 43.7 34.6 – cover Conifer

Coast and Cascades Eastern Cascades Coast Range California Cascades California Klamaths Coast Table C-1–Results MaxEnt the of bootstrappedTable replicate habitat suitability replicates models each modeling (10 for showing region) mean percentage environmental for variable model contributions and changes in model test gain associated with specific of inclusion exclusion or environmental variables Modeling region Washington Washington Oregon Oregon and Oregon and California Note: MaxEnt replicate variable response curve information for each modeling region is available upon request. d.b.h. = diameter at breast height.

107 GENERAL TECHNICAL REPORT PNW-GTR-850

The following sections in this appendix summarize the Wildfire frequency is very low. Federally managed lands MaxEnt modeling regions and the modeling results of the occupy the interior half of the province, the core being bootstrapped replicates. Model region descriptions are Olympic National Park girded by the Olympic National largely based on information from the Landscape Ecology Forest. Most of the Western Lowlands are in private and Modeling, Mapping and Analysis Web site (http://www.fsl. state ownership, with extensive urban and agricultural orst.edu/lemma/main.php?project=nwfp&id=home). areas. It is dominated by wide, glaciated valleys, except for Washington Coast and Cascades Modeling Region the Willapa Hills in the coastal section. Lowland coniferous (MR 221) forest, deciduous forest, and native prairie were its natural dominant vegetation types. The Western Cascades lower This modeling region conforms to the Washington Douglas- elevation forests are dominated by Douglas-fir Pseudotsuga( fir ecological region used in the demographic meta-anal- menziesii (Mirb.) Franco) and western hemlock (Tsuga yses, and contains the Olympic Peninsula and the Rainier heterophylla (Raf.) Sarg.) grading into Pacific silver firAb ( - demographic study areas. It encompasses the Washington ies procera Rehd.) at midelevations, and mountain hemlock Olympic Peninsula, Washington Western Lowlands, and (Tsuga mertensiana (Bong.) Carr.) and subalpine vegetation the Washington Western Cascades physiographic provinces. at higher elevations. Wildfire frequencies are low to moder- The Olympic Peninsula is dominated by moist, produc- ate. About two-thirds of the province is administered by tive coniferous rain forest on the western slope, and drier federal agencies. Douglas-fir forest in the rain shadow on the eastern slope.

Figure C-1–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the Washington Coast and Cascades modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-2 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

108 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-2–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the Washington Coast and Cascades modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat. P/E = predicted vs. expected.

Figure C-3–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for the Washington Coast and Cascades modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95-percent confidence intervals. This histogram shows net change (losses and gains) between 1996 and 2006.

109 GENERAL TECHNICAL REPORT PNW-GTR-850

Washington Eastern Cascades Modeling Region Laws.) forest at lower to midelevations, and by true fir (MR 222) (Abies spp.) and mountain hemlock at higher elevations. This modeling region conforms to the Washington Mixed- Forest productivity is low in places owing to poor soils and Conifer ecological region used in the demographic meta- high elevations. Historically, fire frequencies were high analyses, and contains the Cle Elum demographic study (≤35-year fire-return intervals). Intensive fire suppression th area. It also conforms to the Washington Eastern Cascades practices since the latter half of the 20 century have physiographic province. The slopes of the Washington resulted in areas with significant accumulations of fuel and Eastern Cascades province are dominated by mixed-conifer shifts in species composition and stand structure. About forest and ponderosa pine (Pinus ponderosa Dougl. ex two-thirds of the area is federally managed.

Figure C-4–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the Washington Eastern Cascades modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-5 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

110 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-5–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the Washington Eastern Cascades modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat. P/E = predicted/expected ratio.

Figure C-6–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for the Washington Eastern Cascades modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95- percent confidence intervals. This histogram shows net change (losses and gains) between 1996 and 2006.

111 GENERAL TECHNICAL REPORT PNW-GTR-850

Oregon Coast Range Modeling Region (MR 223) hemlock, and western redcedar (Thuja plicata Donn ex D. This modeling region conforms to the Oregon Coastal Don). The Forest Service and Bureau of Land Management Douglas-fir ecological region used in the demographic together manage about one-quarter of the land in the region. meta-analyses, and contains the Oregon Coast Ranges and Older forests are highly fragmented, largely as a result of Tyee demographic study areas. It contains the Oregon Coast infrequent but very large wildfires in the 1800s and 1900s, physiographic province, and also the Willamette Valley and heavy cutting, as well as checkerboard ownership physiographic province west of the Willamette River, as patterns. Most of the Willamette Valley is in private owner- well as the coastal margins of the Oregon Klamath phys- ship and includes extensive urban and agricultural areas. iographic province. The moist, productive forests in this Lowland coniferous forest, deciduous forest, and native modeling region are dominated by Douglas-fir, western prairie were the natural dominant vegetation types.

Figure C-7–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the Oregon Coast Range modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-8 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

112 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-8–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the Oregon Coast Range modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat.

Figure C-9–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for the Oregon Coast Range modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95-percent confidence intervals. This histogram shows net change (losses and gains) between 1996 and 2006.

113 GENERAL TECHNICAL REPORT PNW-GTR-850

Oregon and California Cascades Modeling Region Protection Agency). This thin delineation represents the (MR 224) ecotone between the East and West Cascades, and not the This modeling region conforms to the Oregon Cascades entire East Cascades province. On the west slope, Douglas- Douglas-fir ecological region used in the meta-analyses, fir and western hemlock give way to Pacific silver fir at and contains the H.J. Andrews and South Cascades midelevations, and mountain hemlock and subalpine fir demographic study areas. It encompasses the east and (Abies lasiocarpa (Hook.) Nutt.) at high elevations. The east west Cascades provinces in Oregon and portions of the slope is covered by mixed-conifer and ponderosa pine forest California Cascades province as delineated along level III at lower elevations, and true firs and mountain hemlock at ecoregion lines. Although there are differences between higher elevations. The southern portion is mixed-conifer the east and west Cascades, our decision to lump them into and pine forests in fire-adapted landscapes. Fire frequencies one modeling region was based on how the east Cascades range from low to high along a north-to-south moisture province was originally drawn to define the eastern margin gradient. Fire suppression has resulted in shifts in species of the owl’s range, which extends into the larger eastern composition and stand structure. About two-thirds of the Cascades ecoregion (as delineated by the Environmental land is administered by federal agencies.

Figure C-10–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the Oregon and California Cascades modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-11 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

114 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-11–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the Oregon and California Cascades modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat.

Figure C-12–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for the Oregon and California Cascades modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95-percent confidence intervals. This histogram shows net change (losses and gains) between 1994/96 and 2006/07.

115 GENERAL TECHNICAL REPORT PNW-GTR-850

Oregon and California Klamaths Modeling Region forest such as Douglas-fir mixed with tanoak Lithocarpus( (MR 225) densiflorus (Hook. & Arn.) Rehd.), and Pacific madrone This modeling region conforms to the Oregon/California (Arbutus menziesii Pursh). The region is characterized Mixed-Conifer ecological region used in the demographic by historically high fire frequencies (≤35-year fire-return meta-analyses, and contains the Klamath and Northwest intervals), and fire suppression has resulted in areas with California demographic study areas. It encompasses the significant accumulations of fuel, shifts in species composi- Klamath physiographic provinces of Oregon and California. tion, and changes in stand structure. Forests are highly It is influenced by unique geologic conditions. In many fragmented as a result of dry climate, poor soils, and past areas, serpentine soils formed by the accretion of rocks onto harvest practices, as well as ownership patterns, especially the continent control the native vegetation, which is domi- in areas of “checkerboard” ownership. Slightly over half of nated by mixed-conifer and mixed-conifer and hardwood the province in Oregon is federally managed. In California, national forests cover about three-quarters of the region.

Figure C-13–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the Oregon and California Klamaths modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-14 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

116 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-14–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the Oregon and California Klamaths modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat.

Figure C-15–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for Oregon and Cali- fornia Klamaths modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95-percent confidence intervals. This histogram shows net change (losses and gains) between 1994/96 and 2006/07.

117 GENERAL TECHNICAL REPORT PNW-GTR-850

California Coast Modeling Region (MR 226) physiographic province to encompass the coastal redwood This modeling region conforms to the California Coast forests in that area. Moist, productive forests in the Califor- ecological region used in demographic meta-analyses, and nia Coast region are dominated by Douglas-fir and western contains the independently operated Green Diamond Re- hemlock, and contain most of the coastal redwood forests. sources and Hoopa Reservation demographic study areas. The southeastern portion of this modeling region falls It conforms to the California Coast Range physiographic within the Central California Chaparral and Oak Wood- province, extending slightly into coastal Oregon Klamath lands ecoregion. Only a small proportion of the California Coast region is administered by federal agencies.

Figure C-16–The mean predicted vs. expected curve (solid black line) from the model replicates, showing 95-percent confidence intervals (gray-shaded vertical bars) for the California Coast modeling region. The logistic thresholds used to define the four-class habitat map are represented by vertical blue-dashed lines. The P/E = 1 threshold is where the curve crosses the random chance line (red-dashed line). The solid black dot represents the 10-percentile threshold (see fig. C-17 below) indicating where 90 percent of the training data (owl pair site centers) occured above that threshold. The mean Spearman rank correlation (Rs) is shown in the upper right-hand corner. See Hirzel et al. (2006) for more information.

118 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure C-17–Habitat modeling statistics produced during the MaxEnt model bootstrapped replicates for the California Coast modeling region. Bars represent the mean statistic value, and error bars show the 95-percent confidence intervals. The first four bars represent model fit and discrimination statistics; the last four bars are common “thresholds” used to classify continuous habitat suitability models into binary maps of “not-suitable” and “suitable” habitat.

Figure C-18–Bookend habitat model area of suitable nesting/roosting habitat for northern spotted owls for the California Coast modeling region. The bars represent the mean estimate of suitable habitat, and error bars show the 95-percent confidence intervals. This histogram shows net change (losses and gains) between 1994 and 2007.

References Hirzel, A.H.; Le Lay, G.; Helfer, V.; Randin, C.; Guisan, A. 2006. Evaluating the ability of habitat suitability models to predict species presences. Ecological Modelling. 199: 142–152.

119 GENERAL TECHNICAL REPORT PNW-GTR-850 Jason Mowdy

120 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix D: Nesting/Roosting Habitat Status and Trend Tables Based on LandTrendr Analysis 0.0 -0.1 -1.6 -0.3 -4.6 -1.2 -0.3 -2.1 -6.2 -4.9 -1.7 -6.6 -1.5 -5.6 -3.7 -13.6 Percent Change

700 24,300 729,000 462,400 494,400 549,400 248,500 132,900 910,900 101,700 2006/07 1,031,600 2,247,300 1,275,200 2,568,200 1,145,500 5,961,000

0 loss -500 -400 Gross -3,500 -1,300 -2,300 -1,600 -22,500 -26,900 -27,000 -86,800 -16,300 -64,600 -68,500 -131,400 -226,800

0 0 0 -200 -700 -800 Wildfire -1,900 -15,900 -16,800 -24,500 -85,000 -13,300 -61,700 -64,400 -122,800 -204,000

Acres – Acres

0 0 0 0 0 -400 -600 -200 -100 disease -1,800 -2,200 -1,700 -2,500 -1,500 -1,600 -6,300 Insects and

-300 -400 -300 -800 -2,400 -4,800 -7,900 -1,300 -1,900 -1,600 -1,300 -6,100 -1,400 -2,500 Harvest -16,500

700

24,700 1994/96 729,500 484,900 495,700 636,200 264,800 135,200 975,500 103,300 – 1,035,100 2,274,200 1,302,200 2,699,600 1,214,000 6,187,800

Estimates nesting/roosting of habitat and loss federal on reserved change- lands using LandTrendr

–

State total State total State total Range total Olympic Peninsula Olympic Lowlands Western Cascades Western Eastern Cascades Coast Range Valley Willamette Cascades Western Klamath Eastern Cascades Coast Range Klamath Cascades detection data detection Table D-1 Physiographic province Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100

121 GENERAL TECHNICAL REPORT PNW-GTR-850 0.0 -0.6 -0.5 -4.0 -1.9 -1.7 -3.7 -1.8 -4.0 -7.0 -2.7 -1.9 -2.5 -6.4 -3.2 -2.7 Percent Change

0 2,600 33,400 10,100 113,400 246,600 181,100 461,100 939,600 334,900 128,400 501,200 102,900 614,200 2006/07 1,518,900 2,594,200

0 loss -200 -100 -200 Gross -1,300 -7,600 -9,100 -2,000 -9,600 -7,000 -16,900 -13,900 -42,500 -13,000 -20,200 -71,800

0 0 0 0 0 -200 Wildfire -4,100 -4,100 -4,400 -8,600 -4,500 -9,900 -1,000 -11,100 -17,500 -32,700

Acres – Acres

0 0 0 0 0 0 -200 -200 -500 -100 -600 -100 -300 -400 disease -1,200 -1,800 Insects and

0 0

-200 -100 -1,300 -3,300 -4,800 -2,000 -5,200 -4,500 -3,000 -5,700 -8,700 Harvest -12,000 -23,800 -37,300

2,700 33,600 10,300 115,400 1994/96 247,900 188,700 470,200 956,500 348,800 138,000 514,200 109,900 634,400 – 1,561,400 2,666,000

Estimates nesting/roosting of habitat and loss federal on nonreserved change- lands using LandTrendr

–

State total State total State total Range total Western Lowlands Olympic Peninsula Western Cascades Western Eastern Cascades Coast Range Valley Willamette Cascades Western Klamath Eastern Cascades Coast Range Klamath Cascades Table D-2 data detection Physiographic province Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100

122 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl -0.1 -1.6 -0.4 -4.5 -1.3 -0.5 -2.9 -1.9 -6.4 -4.1 -1.7 -5.2 -4.0 -4.8 -3.4 -10.2 Percent Change

3,300 24,300 762,400 643,500 607,900 884,300 377,000 142,900 204,600 2006/07 1,278,200 2,708,400 2,214,800 4,087,300 1,412,200 1,759,700 8,555,400

loss -700 -400 -100 Gross -4,800 -3,300 -2,500 -8,600 -30,100 -36,000 -43,900 -25,900 -77,600 -88,700 -100,700 -173,900 -298,600

0 0 0 -200 -700 Wildfire -2,100 -1,800 -20,000 -20,900 -28,900 -93,600 -17,800 -71,600 -75,500 -140,300 -236,700

Acres – Acres

0 0 0 0 -400 -300 -100 -300 disease -2,000 -2,400 -1,100 -2,300 -3,700 -1,600 -2,000 -8,100 Insects and

-500 -400 -100 -300 -3,700 -8,100 -3,300 -6,800 -5,800 -4,400 -6,500 Harvest -11,200 -12,700 -13,900 -29,900 -53,800

3,400 24,700 611,200 1994/96 763,100 673,600 985,000 402,900 145,400 213,200 – 1,283,000 2,744,400 2,258,700 4,261,200 1,489,800 1,848,400 8,854,000

Estimates nesting/roosting of habitat and loss all on federal change-detection lands using LandTrendr

—

State total State total State total Range total Olympic Peninsula Lowlands Western Cascades Western Eastern Cascades Range Coast Valley Willamette Cascades Western Klamath Eastern Cascades Coast Range Klamath Cascades Table D-3 data Physiographic province Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100

123 GENERAL TECHNICAL REPORT PNW-GTR-850 -1.2 -4.9 -5.6 -3.7 -1.9 -2.7 -3.2 -2.7 -1.3 -4.1 -4.8 -3.4 -6.3 -7.0 -8.6 -5.5 -7.3 -19.3 -22.4 -15.5 Percent Change

461,100 614,200 2006/07 2,247,300 2,568,200 1,145,500 5,961,000 1,518,900 2,594,200 2,708,400 4,087,300 1,759,700 8,555,400 1,016,300 1,073,400 1,459,000 3,548,700 3,724,700 5,160,700 3,218,700 12,104,100

loss -9,100 Gross -26,900 -68,500 -42,500 -20,200 -71,800 -36,000 -88,700 -97,700 -131,400 -226,800 -173,900 -298,600 -242,600 -309,000 -649,300 -278,600 -482,900 -186,400 -947,900

-4,100 -2,400 -5,100 -5,600 -11,100 -16,800 -64,400 -17,500 -32,700 -20,900 -75,500 -13,100 -23,300 -81,100 Wildfire -122,800 -204,000 -140,300 -236,700 -145,400 -249,800

Acres – Acres

-200 -400 -2,200 -2,500 -1,600 -6,300 -1,200 -1,800 -2,400 -3,700 -2,000 -8,100 -6,000 -2,700 -1,900 -8,400 -6,400 -3,900 disease -10,600 -18,700 Insects and

-7,900 -6,100 -2,500 -4,800 -8,700 Harvest -11,200 -16,500 -23,800 -37,300 -12,700 -29,900 -53,800 -90,200 -234,200 -301,200 -625,600 -246,900 -331,100 -101,400 -679,400

470,200 634,400 – 1994/96 2,274,200 2,699,600 1,214,000 6,187,800 1,561,400 2,666,000 2,744,400 4,261,200 1,848,400 8,854,000 1,258,900 1,382,400 1,556,700 4,198,000 4,003,300 5,643,600 3,405,100 13,052,000

Range total Range total Range total Range total Range total Oregon California Oregon California Oregon California Oregon California Oregon California Washington Washington Washington Washington Washington Estimates of nesting/roosting habitat and loss on all habitat-capable lands within the owl’s range using habitat and loss on all habitat-capable lands within the owl’s D-4 – Estimates of nesting/roosting Table change-detection data LandTrendr Land class Federal reserved: Federal nonreserved: All federal: Nonfederal: All lands: Note: Acres are rounded to the nearest acres. 100

124 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix E: Dispersal Habitat Status and Trend Tables Based on LandTrendr Analysis

4.3 7.4 4.4 1.7 3.6 8.7 3.1 2.4 2.0 7.0 2.8 7.0 3.6 3.0 -8.8 11.9 Percent Change

2,500 66,400 830,600 824,100 677,200 181,200 261,600 2006/07 1,098,000 1,972,800 1,453,200 4,590,400 2,025,000 4,359,400 1,994,500 2,437,300 11,387,100

200 Net 4,600 11,900 45,400 83,500 24,800 88,400 61,200 15,700 85,900 54,900 17,200 84,000 change -79,600 158,300 328,200

200 5,100 Gain 46,100 87,700 71,400 90,200 93,400 43,900 44,300 14,600 18,600 210,300 272,000 127,000 160,200 642,500

0 0 -400 -600 -100 -700 -2,300 dispersal habitat federal on reserved lands Wildfire -41,600 -42,600 -29,600 -22,100 -67,600 -70,600 Acres – Acres -121,400 -173,200 -286,400

0 0 0 0 0 -400 -600 -400 -100 disease -1,300 -1,700 -3,100 -4,100 -2,000 -2,100 -7,900 Insects and

-300 -500 -300 -700 -3,200 -3,700 -7,700 -1,700 -2,000 -1,700 -3,400 -8,800 -2,500 -3,500 Harvest -20,000

2,300 61,800 – 742,200 903,700 661,500 169,300 244,400 1994/96 1,052,600 1,889,300 1,428,400 4,432,100 1,963,800 4,273,500 1,939,600 2,353,300 11,058,900

–Estimates gross of loss, gross gain, of and net change

State total State total State total Range total Olympic Peninsula Lowlands Western Cascades Western Eastern Cascades Coast Range Valley Willamette Cascades Western Klamath Eastern Cascades Coast Range Klamath Cascades Table E-1 Physiographic province Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100

125 GENERAL TECHNICAL REPORT PNW-GTR-850

8.1 6.6 8.8 11.8 21.1 50.0 10.7 10.4 16.6 10.1 12.6 10.3 15.5 10.4 10.0 10.2 Percent Change

300 11,400 88,900 23,800 491,300 426,000 304,200 645,100 390,600 363,200 2006/07 1,006,500 1,646,300 2,997,600 1,099,100 1,486,100 5,490,200

100 Net 1,200 3,200 15,500 47,400 32,100 95,100 43,400 40,000 43,600 29,400 change 151,200 279,400 103,100 135,700 510,200

100 1,300 4,100 Gain 15,700 48,900 45,100 45,800 62,200 57,600 43,600 109,800 167,100 334,000 124,200 171,900 615,700

0 0 0 0 0 -800 -9,200 -9,200 -4,200 -6,800 -2,000 dispersal habitat federal on nonreserved lands Wildfire -16,100 -27,100 -16,200 -19,000 -55,300 Acres – Acres

0 0 0 0 0 0 -300 -300 -700 -300 -700 -300 disease -1,700 -4,000 -4,300 -6,300 Insects and

-200 -100 -100 -1,500 -3,500 -5,200 -2,400 -5,800 -6,500 -4,600 -8,200 Harvest -11,000 -25,800 -12,900 -43,900

200

73,400 10,200 20,600 – 911,400 443,900 393,900 260,800 605,100 347,000 996,000 333,800 1994/96 1,495,100 2,718,200 1,350,400 4,980,000

Estimates gross of loss, gross gain, of and net change

–

State total State total State total Range total Western Lowlands Olympic Peninsula Western Cascades Western Eastern Cascades Coast Range Valley Willamette Cascades Western Klamath Eastern Cascades Coast Range Klamath Cascades Table E-2 Physiographic province Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100

126 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl 5.4 7.6 5.6 3.1 4.7 6.1 5.9 5.2 8.0 5.4 8.1 5.9 5.2 -2.6 11.2 13.1 Percent Change

66,700 13,900 205,000 624,800 2006/07 1,186,900 2,464,100 1,879,200 5,596,900 1,134,800 3,671,300 1,469,200 1,067,800 7,357,000 3,093,600 3,923,400 16,877,300

Net 4,700 1,400 60,900 56,900 59,300 15,100 46,600 change -39,600 130,900 253,400 131,800 212,400 365,300 158,000 219,700 838,400

5,200 1,500 Gain 61,800 18,700 62,200 116,500 136,600 320,100 136,000 260,500 106,100 101,900 606,000 251,200 332,100 1,258,200

0 0 -400 -600 -100 -3,100 -2,700 dispersal habitat all on federal lands Wildfire -50,800 -51,800 -33,800 -28,900 -83,800 -89,600 Acres – Acres -137,500 -200,300 -341,700

0 0 0 0 -400 -700 -100 disease -1,600 -2,000 -1,300 -3,800 -5,800 -2,300 -4,000 -6,400 -14,200 Insects and

-500 -500 -100 -400 -4,700 -7,200 -4,100 -7,500 -9,900 -7,100 -8,900 Harvest -12,900 -13,000 -34,600 -16,400 -63,900

62,000 12,500 – 189,900 578,200 1994/96 1,126,000 2,333,200 1,822,300 5,343,500 1,003,000 3,458,900 1,508,800 1,008,500 6,991,700 2,935,600 3,703,700 16,038,900

–Estimates gross of loss, gross gain, of and net change

State total State total State total Range total Olympic Peninsula Lowlands Western Cascades Cascades Western Eastern Coast Range Valley Willamette Cascades Western Cascades Klamath Eastern Coast Range Klamath Cascades Washington: Oregon: California: Note: Acres are rounded to the nearest acres. 100 Table E-3 Physiographic province

127 GENERAL TECHNICAL REPORT PNW-GTR-850

3.6 2.0 3.6 3.0 4.7 5.2 5.9 5.2 6.5 4.7 9.2 5.5 5.0 6.9 Net 10.4 10.3 10.0 10.2 19.8 12.0 change Percent

2006/07 4,590,400 4,359,400 2,437,300 1,006,500 2,997,600 1,486,100 5,490,200 5,596,900 7,357,000 3,923,400 4,641,100 4,322,700 3,423,600 7,347,000 11,387,100 11,679,700 16,877,300 12,387,400 10,238,000 29,264,700

gain Gross 210,300 272,000 160,200 642,500 109,800 334,000 171,900 615,700 320,100 606,000 332,100 993,000 971,200 443,900 776,000 1,258,200 2,408,100 1,313,100 1,577,200 3,666,300

8,400 loss Gross -52,000 -76,200 -14,700 -54,600 -36,200 -66,700 120,800 -112,400 -711,000 -186,100 -314,300 -105,500 -240,700 -419,800 -777,900 -777,700 -1,368,100 -1,018,600 -1,787,900

9,900 -9,200 -7,000 -7,100 -42,600 -70,600 -27,100 -19,000 -55,300 -51,800 -89,600 -10,000 -58,800 -79,700 dispersal habitat all on habitat-capable lands within Wildfire -173,200 -286,400 -200,300 -341,700 -210,300 -348,800 Acres – Acres

-300 2,900 -1,700 -4,100 -2,100 -7,900 -1,700 -4,300 -6,300 -2,000 -5,800 -6,400 -7,200 -3,500 disease -14,200 -14,700 -19,000 -16,700 -13,000 -33,200 Insects and

-7,700 -8,800 -3,500 -5,200 91,600 Harvest -20,000 -25,800 -12,900 -43,900 -12,900 -34,600 -16,400 -63,900 108,000 -689,300 -760,700 -702,200 -795,300 -1,342,000 -1,405,900

– 911,400 1994/96 4,432,100 4,273,500 2,353,300 2,718,200 1,350,400 4,980,000 5,343,500 6,991,700 3,703,700 4,359,100 4,129,400 2,858,900 9,702,600 6,562,600 11,058,900 11,347,400 11,121,100 16,038,900 27,386,300

Estimates gross of loss, gross gain, of and net change

–

Range total Range total Range total Range total Range total Washington Oregon Washington Oregon California Washington Oregon California Washington Oregon California Washington Oregon California California nonreserved: Table E-4 range owl’s the Land class Federal reserved: Federal All federal: Nonfederal: All lands: Note: Acres are rounded to the nearest acres. 100

128 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix F: Crosswalk for Modifying Bookend 2 (2006/07) Map for Making Habitat Suitability Histograms

As discussed in chapter 3, we chose to not report on the between bookend 1 (1994/96) and bookend 2. This approach highly suspect suitable habitat gains in nesting/roosting is similar to what was done in the 10-year report, where habitat for this monitoring cycle given the short timespan we had only a baseline map and change-detection data to of our analysis and because of uncertainties with model estimate habitat changes. The habitat suitability histograms transferability, bookend 2 (2006/07) map reviews with on the following pages are formatted to be similar to the 1-m color aerial imagery Natonal Agricultural Imagery histograms in appendix G of the 10-year report. The modi- Program, geographic information system (GIS) analysis fied bookend 2 map (as described below) is our best esti- of model variable changes, and an inventory plot analysis mate of habitat classes as of 2006/07. It is conservative in (app. H). However, we anticipate that our ability to measure nature, as it maintains suitable habitat classes (3 and 4) from these gains will improve with the passing of more time to bookend 1, and only shows loss in these suitable classes if separate the bookend maps, and with improved remote sens- verified by LandTrendr (LT) data. We allow for minor shifts ing technologies. This appendix presents the following table within habitat classes that may represent subtle changes but to describe the crosswalk we used for creating a modified do not result in a change between the broader categories of bookend 2 map for the purpose of making “habitat suit- “unsuitable/marginal” (i.e., classes 1 and 2) and the “suitability histograms” to help visualize shifts in habitat classes able” classes (i.e., classes 3 and 4).

Table F-1–Crosswalk table for modified bookend 2 map Bookend model Modified 2006/07 habitat classes Assumptions (based on aerial image review habitat class 1994/6 2006/7 and GIS analysis of environmental variables) (modified bookend 2) 1 1 Not suitable either period; no change 1 1 2 Not suitable either period; accept shift from unsuitable to marginal 2 1 3 Trend toward suitable, but highly uncertain gain; limit shift to marginal class 2 1 4 Trend toward suitable, but highly uncertain gain; limit shift to marginal class 2 2 1 Not suitable either period; accept shift from marginal to unsuitable 1 2 2 Not suitable either period; no change 2 2 3 Trend toward suitable, but highly uncertain gain; keep in marginal class 2 2 4 Trend toward suitable, but highly uncertain gain; keep in marginal class 2 3 1 If LTa verified habitat loss, moved to unsuitable; otherwise no change 1 if LT verified, otherwise 3 3 2 If LT verified habitat loss, moved to marginal; otherwise no change 2 if LT verified, otherwise 3 3 3 Suitable habitat both periods; no change 3 3 4 Suitable habitat both periods; accept shift to highly suitable 4 4 1 If LT verified habitat loss, moved to unsuitable; otherwise no change 1 if LT verified, otherwise 4 4 2 If LT verified habitat loss, moved to marginal; otherwise no change 2 if LT verified, otherwise 4 4 3 Suitable habitat both periods; accept shift to suitable (degraded) 3 4 4 Suitable habitat both periods; no change 4 a LT = LandTrendr. GIS = geographic information system.

129 GENERAL TECHNICAL REPORT PNW-GTR-850

The histograms on the following pages establish the as described in chapter 3. The first bar in the pair shows format we propose for visually representing status and conditions at time 1 (1994/96); the second bar shows condi- trends in habitat classes for future monitoring efforts. They tions at time 2 (2006/07), based on our modified bookend are based on habitat conditions at roughly the time of the 2 map. They also provide a visual on how owl habitat is Northwest Forest Plan (the Plan) implementation (1994/96) distributed across reserved and nonreserved federal lands. to the end of our analysis data set in 2006/07. There are four The example histogram below is provided to help interpret pairs of histogram bars, one pair per habitat suitability class the histograms provided for each physiographic province in the following pages.

In the example above, we observe a slight decrease interpretation indicates that the increase in the marginal (3.5 percent) in unsuitable habitat class 1 between time 1 habitat class (class 2) will continue to progress to the suit- and time 2. We also observe a slightly larger (4.4 percent) able classes with time. This province has very little suitable increase in marginal habitat class (but still unsuitable for nesting/roosting habitat. nesting/roosting). There has been a slight decrease in the The table under the graphs shows the estimates of suitable habitat class 3 (1.1 percent) with a very slight (0.2 percentage of forest-capable land changes between habitat percent) increase in the highly suitable habitat class. We can maps for both periods. The percentages are split into conclude that forest succession in habitat class 1 accounted nonreserved and reserved land use allocations. The fol- for most of the increase in habitat class 2, but some of the lowing graphs illustrate our best estimate of how habitat loss in habitat class 3 may have also accounted for some is changing (trending) at this early stage of the Plan. of the changes in habitat class 2, or perhaps offset some These graphs are primarily for interpretive purposes. The of the decrease in habitat class 1. The slight increase in observed change between the bookends is small, with the habitat class 4 may be a result of changes in habitat class 3, largest changes being increases in the marginal classes. We as seen in the plot analysis (app. H) where there may have observed similar changes in dispersal habitat (see chapter 3) been some subtle changes. However, this does not result and consider this an indication of noticeable future recruit- in a change in the broader “suitable” class. The simplest ment from marginal to suitable habitat within the next two to three decades.

130 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

131 GENERAL TECHNICAL REPORT PNW-GTR-850

132 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

133 GENERAL TECHNICAL REPORT PNW-GTR-850 Ray Davis

134 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix G: Wildfire Suitability Modeling, MaxEnt Replicate Data

Table G-1–Wildfire suitabilty modeling environmental variables and their model contributions Environmental Contribution variable Description Value range Units to model Percent August maximum PRISM (1971–2000)—mean maximum 1,456–3,635 °C (x 100) 27.8 temperature temperature for month of August Lightning ignition Kernel density map of all lightning- 0–992 Ignitions/km² density caused fire ignitions 1970–2002, (x100) 24.1 from Brown et al. (2002)a Slope Percentage of slope based on analysis 0–125 Percent 23.7 of Digital Elevation Model Distance from road Linear distance to nearest road, based 0–28,300 Meters 14.6 on road layer in Gallo et al. (2011)b Summer precipitation PRISM (1971–2000)—mean rainfall 3,243–6,884 ln mm 4.4 between May and September, log (x1,000) transformed Elevation U.S. Geological Survey 0–2477 Meters 4.2 Digital Elevation Model Solar radiation Potential relative solar radiation 5,619–20,546 Index 1.1 as derived by Pierce et al. (2005)c a Brown, T.J.; Hall, B.L.; Mohrle, C.R.; Reinbold, H.J. 2002. b Gallo, K.; Lanigan, S.H.; Eldred, P.; Gordon, S.N.; Moyer, C. 2005. c Pierce, K.B.; Lookingbill, T.R.; Urban, D.L. 2005.

Figure G-1–Correlation matrix (Pearson correlations) for environmental variables used in the model.

135 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure G-2–Averaged model fit, accuracy, and threshold statistics (with 95-percent confidence intervals) from the 10 bootstrapped model replicates. Note that the predicted vs expected (P/E) = 1 threshold is similar to the maximum testing sensitivity plus specificity threshold, which minimizes model omission and commission errors.

136 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Figure G-3–Model response curves showing logistic probability of large wildfire occurrence (y-axis) for each environmental variable, as it is varied in jackknifed models, keeping all other environmental variables at their average sample values. PRP = potential relative radiation. 137 GENERAL TECHNICAL REPORT PNW-GTR-850

Figure G-4—Jackknife modeling results for variable importance. Note the similarities between the regularized training gain (top) and test gain (middle) graphs. The high level of similarity between them indicates that the model is not over-fit.

138 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Literature Cited Gallo, K.; Lanigan, S.H.; Eldred, P.; Gordon, S.N.; Brown, T.J.; Hall, B.L.; Mohrle, C.R.; Reinbold, Moyer, C. 2005. Northwest Forest Plan—the first H.J. 2002. Coarse assessment of federal wildland 10 years (1994–2003): preliminary assessment of fire occurrence data: report for the National Wildfire the condition of watersheds. Gen. Tech. Rep. PNW- Coordinating Group. Reno, NV: CEFA Report 02-04. GTR-647. Portland, OR: U.S. Department of Agricul- Program for Climate, Ecosystem and Fire Applications, ture, Forest Service, Pacific Northwest Research Station. Desert Research Institute, Division of Atmospheric 133 p. Sciences. 31 p. Pierce, K.B.; Lookingbill, T.R.; Urban, D.L. 2005. A simple method for estimating potential relative radiation (PRR) for landscape-scale vegetation analysis. Landscape Ecology. 20: 137–147 Ray Davis

139 GENERAL TECHNICAL REPORT PNW-GTR-850 Ray Davis

140 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Appendix H: Regional Inventory Plot Analysis Carol A. Applea and Raymond J. Davisb

PLOT ANALYSIS • Top story quadratic mean diameter for only conifer An analysis of regional vegetation inventory plots was and not all species performed to determine if there have been significant gains • Use of a different strata attribute in northern spotted owl (Strix occidentalis caurina) nesting/ These differences were made to adjust to the new roosting habitat since monitoring was implemented. We gradient nearest neighbor variables used for the spatial used plot data from the Current Vegetation Survey (CVS) habitat modeling (see chapter 3, this report). The intent of inventory program on USDA Forest Service (FS) and USDI this analysis was to explore for differences in “query group” Bureau of Land Management (BLM) lands in Oregon and acres between the initial measurement and remeasurement Washington. Forest Inventory and Analysis (FIA) program period, and not for differences between the 10-year report data were used for FS lands in California. Data were not map results and this report’s map results. In addition, the available for USDI National Park Service lands or BLM results of this analysis should not be compared to the results lands in Washington or California. of the 10-year report plot analysis. The CVS inventory provides comprehensive informa- On Pacific Northwest Region FS lands, nearly all CVS tion on vegetative resources on FS lands in Oregon and plots have had two samples, but on Oregon BLM lands, only Washington and BLM lands in the Northwest Forest Plan one-quarter of the original CVS plots had been remeasured area in Oregon. The CVS plots consist of four grids of field at the time of this analysis. Based on the numbers of plots plots that are separated by 3.4 mi on a north-south, east- for each year of measurement, the weighted average year for west direction. These four inventory grids are offset from initial plot measurements in Oregon and Washington was one another to produce one single 1.7-mi grid of plots across 1995 for initial plot measurements and 2002 for remeasure- BLM lands and FS lands, except in wilderness areas where ments. For California, the plot measurement period spans the grid density is 3.4 mi. The FIA plots for FS lands in 1997 to 2005. The first inventory in California was con- California are also distributed geographically on a 3.4-mi ducted under the older FIA “periodic” sample design. This grid. For specific information on the attributes that are protocol was replaced by FIA’s “annual” sample design, collected on FS lands, refer to the Web sites: http://www. which was used for the plot remeasurments. This change in fs.fed.us/r6/survey/ and http://www.fs.fed.us/r5/rsl/projects/ inventory protocol confounds inferences on habitat changes inventory/InvInfo.shtml. Refer to pages 31–36 in Moeur et in California, as “real” change cannot be separated from al. (2005) for additional discussion of the CVS and FIA. effects related to changing sampling protocols (see Moeur et A spotted owl nesting/roosting habitat query was al. 2011) for more discussion. developed (table H-1) that is similar to what was used in the As in the 10-year report, the “query groups” in table 10-year report (Davis and Lint 2005). There were differ- H-1 represent a progression of stand conditions, based on ences that included: conifer diameter, total canopy cover, and stand structure • Use of a slightly different set of plots complexity (strata) that represent habitat similarity to condi- • Summarized data at the plot level vs. subplot level tions used by spotted owls for nesting and roosting. A query was applied to both the initial measurement and remeasure- a Carol A. Apple, Mathematical statistician, Regional Vegeta- ment plot data in each physiographic province that occurred tion Inventory Program, U.S. Department of Agriculture, Forest Service, Pacific Northwest Region, 333 SW First Ave., Portland, within the “habitat capable” areas described in Davis and OR 97204. Lint (2005) to assign a group code to each plot. In addition b Raymond J. Davis, Northern Spotted Owl Monitoring Module to the six groups in table H-1, two combined groups of EF Leader, Northwest Forest Plan Interagency Monitoring Program, U.S. Department of Agriculture, Forest Service, Pacific Northwest and DEF were also assigned. Region, 2900 NW Stewart Parkway, Roseburg, OR 97471.

141 GENERAL TECHNICAL REPORT PNW-GTR-850

Table H-1—Forest stand condition query for Current Vegetation Survey plot data Low <------Spotted owl nesting/roosting habitat similarity ------> High Query group Query attributes A B C D E F

Query part 1: Quadratic mean diameter (inches) <10.5 ≥10.5 10.5–20.5 10.5–20.5 ≥20.5 ≥20.5 Canopy cover (percent) All ≤40 >40 >40 41–70 >70 Strata All All 1 ≥2 ≥2 ≥2

Query part 2: Quadratic mean diameter (inches) NA NA NA ≥20.5 NA NA Canopy cover (percent) >40 Strata 1

An analysis was then done using the jackknife method class E, but these were concurrent with significant increases to estimate the variance of mean acres for each query group in class F; therefore, these changes “cancelled” each other by measurement period and province. The variance was out, resulting in no significant change in the EF group. used in performing a t-test to look at the differences in the Significant, decreases in EF and DEF were observed in means between the two periods. This test assumes indepen- the California Cascades and Klamath Mountain provinces dence between the two samples, but in reality many of the (table H-4); however, as stated above, the change in protocol plots were remeasured. Taking that into consideration, this used to collect plot information in California confounds this test provided conservative results for significance: if it is inference. The histograms on the following page display the significant, it is very significant. results of the plot analysis for each physiographic province The results of this analysis did not show any evidence with significant amounts of federal lands. The results of the of significant habitat recruitment into classes EF or the t-test are shown in tables H-2 thru H-4. broader class DEF. There were some significant decreases in

142 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl Figure H-1—Habitat changes between measurement plot periods and to (Time 1997 Time 2 = 2002 1 = 1995 to 2005). Histrograms represent estimate mean of acres with 90-percent confidence intervalsfor each period.

143 GENERAL TECHNICAL REPORT PNW-GTR-850

Table H-2–Test for significant difference (bold-faced) of mean acres between measurement periods by physiographic province by query group at the 0.1 significance level (t = 1.6448), Washington Physiographic province Query group t-value Net change Acres Olympic Peninsula A 0.386 -9,300 B 0.379 -1,900 C 0.197 -1,500 D 1.016 24,000 E 1.455 -15,100 F 0.157 3,900 DEF 0.442 12,700 EF 0.438 -11,200 Western Cascades A 1.346 -65,000 B 1.427 -31,200 C 0.559 6,700 D 0.824 44,800 E 2.057 -80,000 F 2.134 115,600 DEF 1.143 80,400 EF 0.562 35,600 Eastern Cascades A 3.584 151,200 B 2.367 -85,000 C 0.337 -1,900 D 1.738 -75,400 E 0.477 14,100 F 0.114 -1,600 DEF 1.254 -63,000 EF 0.384 12,400

144 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

Table H-3–Test for significant difference (bold-faced) of mean acres between measurement periods by physiographic province by query group at the 0.1 significance level (t = 1.6448), Oregon Physiographic province Query group t-value Net change Acres Coast Range A 1.206 -46,500 B 0.061 900 C 0.295 6,800 D 0.770 -37,600 E 0.318 11,200 F 1.527 52,400 DEF 0.479 26,000 EF 1.374 63,600

Western Cascades A 0.480 30,500 B 1.981 -85,000 C 0.472 -14,500 D 0.945 85,400 E 2.988 -175,300 F 2.993 140,500 DEF 0.526 50,500 EF 0.499 -34,800

Klamath A 1.400 -70,500 B 0.287 7,600 C 0.854 8,400 D 1.758 98,000 E 1.478 -77,700 F 1.281 40,800 DEF 0.924 61,000 EF 0.638 -36,900

Eastern Cascades A 1.290 44,500 B 1.743 -45,700 C 0.272 -5,200 D 0.055 2,900 E 0.034 700 F 0.740 11,300 DEF 0.272 14,900 EF 0.464 12,000

145 GENERAL TECHNICAL REPORT PNW-GTR-850

Table H-4—Test for significant difference (bold-faced) of mean acres between measurement periods by physiographic province by query group at the 0.1 significance level (t = 1.6448), California Physiographic province Query group t-value Net change Acres Coast Range A 0.128 1,900 B 1.145 12,300 C — — D 0.309 -4,200 E 0.229 -2,300 F 0.213 2,400 DEF 0.269 -4,100 EF 0.008 100

Klamath A 2.988 259,400 B 1.404 108,700 C 1.722 25,200 D 0.597 63,700 E 0.901 -78,800 F 4.835 -402,400 DEF 3.887 -417,500 EF 4.496 -481,200

Cascades A 0.929 43,400 B 0.968 46,000 C 0.271 3,500 D 0.994 -57,600 E 1.660 -57,000 F 1.681 -36,400 DEF 2.628 -151,000 EF 2.380 -93,500

146 Northwest Forest Plan—the First 15 Years (1994–2008): Status and Trends of Northern Spotted Owl

References Moeur, M.; Ohmann, J.L.; Kennedy, R.E.; Cohen, W.B.; Davis, R.; Lint, J. 2005. Habitat status and trend. In: Gregory, M.J.; Yang, Z.; Roberts, H.M.; Spies, T.A.; Lint, J., tech. coord. Northwest Forest Plan—the first Fiorella, M. [In press.] Northwest Forest Plan—status 10 years (1994–2003): status and trends of northern and trends of late-successional and old-growth forests spotted owl populations and habitat. Gen. Tech. Rep. from 1994 to 2007. Gen. Tech. Rep. Portland, OR: U.S. PNW-GTR-648. Portland, OR: U.S. Department of Department of Agriculture, Forest Service, Pacific Agriculture, Forest Service, Pacific Northwest Research Northwest Research Station. Station: 21–82. Moeur, M.; Spies, T.A.; Hemstrom, M.; Alegria, J.; Browning, J.; Cissel, J.; Cohen, W.B.; Demeo, T.E.; Healey, S.; Warbington, R. 2005. Northwest Forest Plan—the first 10 years (1994–2003): status and trend of late-successional and old-growth forest. Gen. Tech. Rep. PNW-GTR-646. Portland, OR: U.S. Department of Agriculture, Forest Service, Pacific Northwest Research Station. 142 p. Jason Mowdy

147 Pacific Northwest Research Station Web site http://www.fs.fed.us/pnw Telephone (503) 808-2592 Publication requests (503) 808-2138 FAX (503) 808-2130 E-mail [email protected] Mailing address Publications Distribution Pacific Northwest Research Station P.O. Box 3890 Portland, OR 97208-3890 U.S. Department of Agriculture Pacific Northwest Research Station 333 SW First Avenue P.O. Box 3890 Portland, OR 97208-3890 Official Business Penalty for Private Use, $300